prometheus/storage/remote/queue_manager.go

1413 lines
44 KiB
Go
Raw Normal View History

2017-05-10 02:44:13 -07:00
// Copyright 2013 The Prometheus Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package remote
import (
"context"
"errors"
2017-05-10 02:44:13 -07:00
"math"
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
"strconv"
2017-05-10 02:44:13 -07:00
"sync"
"time"
"github.com/go-kit/log"
"github.com/go-kit/log/level"
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
"github.com/gogo/protobuf/proto"
"github.com/golang/snappy"
"github.com/opentracing/opentracing-go"
"github.com/opentracing/opentracing-go/ext"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/common/model"
"go.uber.org/atomic"
2017-05-10 02:44:13 -07:00
"github.com/prometheus/prometheus/config"
"github.com/prometheus/prometheus/model/labels"
"github.com/prometheus/prometheus/model/relabel"
2017-10-23 13:28:17 -07:00
"github.com/prometheus/prometheus/prompb"
"github.com/prometheus/prometheus/scrape"
"github.com/prometheus/prometheus/tsdb/chunks"
"github.com/prometheus/prometheus/tsdb/record"
"github.com/prometheus/prometheus/tsdb/wal"
2017-05-10 02:44:13 -07:00
)
const (
// We track samples in/out and how long pushes take using an Exponentially
// Weighted Moving Average.
ewmaWeight = 0.2
shardUpdateDuration = 10 * time.Second
// Allow 30% too many shards before scaling down.
shardToleranceFraction = 0.3
)
type queueManagerMetrics struct {
reg prometheus.Registerer
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
samplesTotal prometheus.Counter
exemplarsTotal prometheus.Counter
metadataTotal prometheus.Counter
failedSamplesTotal prometheus.Counter
failedExemplarsTotal prometheus.Counter
failedMetadataTotal prometheus.Counter
retriedSamplesTotal prometheus.Counter
retriedExemplarsTotal prometheus.Counter
retriedMetadataTotal prometheus.Counter
droppedSamplesTotal prometheus.Counter
droppedExemplarsTotal prometheus.Counter
enqueueRetriesTotal prometheus.Counter
sentBatchDuration prometheus.Histogram
highestSentTimestamp *maxTimestamp
pendingSamples prometheus.Gauge
pendingExemplars prometheus.Gauge
shardCapacity prometheus.Gauge
numShards prometheus.Gauge
maxNumShards prometheus.Gauge
minNumShards prometheus.Gauge
desiredNumShards prometheus.Gauge
sentBytesTotal prometheus.Counter
metadataBytesTotal prometheus.Counter
maxSamplesPerSend prometheus.Gauge
}
func newQueueManagerMetrics(r prometheus.Registerer, rn, e string) *queueManagerMetrics {
m := &queueManagerMetrics{
reg: r,
}
constLabels := prometheus.Labels{
remoteName: rn,
endpoint: e,
}
m.samplesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "samples_total",
Help: "Total number of samples sent to remote storage.",
ConstLabels: constLabels,
})
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.exemplarsTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "exemplars_total",
Help: "Total number of exemplars sent to remote storage.",
ConstLabels: constLabels,
})
m.metadataTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "metadata_total",
Help: "Total number of metadata entries sent to remote storage.",
ConstLabels: constLabels,
})
m.failedSamplesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "samples_failed_total",
Help: "Total number of samples which failed on send to remote storage, non-recoverable errors.",
ConstLabels: constLabels,
})
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.failedExemplarsTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "exemplars_failed_total",
Help: "Total number of exemplars which failed on send to remote storage, non-recoverable errors.",
ConstLabels: constLabels,
})
m.failedMetadataTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "metadata_failed_total",
Help: "Total number of metadata entries which failed on send to remote storage, non-recoverable errors.",
ConstLabels: constLabels,
})
m.retriedSamplesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "samples_retried_total",
Help: "Total number of samples which failed on send to remote storage but were retried because the send error was recoverable.",
ConstLabels: constLabels,
})
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.retriedExemplarsTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "exemplars_retried_total",
Help: "Total number of exemplars which failed on send to remote storage but were retried because the send error was recoverable.",
ConstLabels: constLabels,
})
m.retriedMetadataTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "metadata_retried_total",
Help: "Total number of metadata entries which failed on send to remote storage but were retried because the send error was recoverable.",
ConstLabels: constLabels,
})
m.droppedSamplesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "samples_dropped_total",
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
Help: "Total number of samples which were dropped after being read from the WAL before being sent via remote write, either via relabelling or unintentionally because of an unknown reference ID.",
ConstLabels: constLabels,
})
m.droppedExemplarsTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "exemplars_dropped_total",
Help: "Total number of exemplars which were dropped after being read from the WAL before being sent via remote write, either via relabelling or unintentionally because of an unknown reference ID.",
ConstLabels: constLabels,
})
m.enqueueRetriesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "enqueue_retries_total",
Help: "Total number of times enqueue has failed because a shards queue was full.",
ConstLabels: constLabels,
})
m.sentBatchDuration = prometheus.NewHistogram(prometheus.HistogramOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "sent_batch_duration_seconds",
Help: "Duration of send calls to the remote storage.",
Buckets: append(prometheus.DefBuckets, 25, 60, 120, 300),
ConstLabels: constLabels,
})
m.highestSentTimestamp = &maxTimestamp{
Gauge: prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "queue_highest_sent_timestamp_seconds",
Help: "Timestamp from a WAL sample, the highest timestamp successfully sent by this queue, in seconds since epoch.",
ConstLabels: constLabels,
}),
}
m.pendingSamples = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "samples_pending",
Help: "The number of samples pending in the queues shards to be sent to the remote storage.",
ConstLabels: constLabels,
})
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.pendingExemplars = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "exemplars_pending",
Help: "The number of exemplars pending in the queues shards to be sent to the remote storage.",
ConstLabels: constLabels,
})
m.shardCapacity = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "shard_capacity",
Help: "The capacity of each shard of the queue used for parallel sending to the remote storage.",
ConstLabels: constLabels,
})
m.numShards = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "shards",
Help: "The number of shards used for parallel sending to the remote storage.",
ConstLabels: constLabels,
})
m.maxNumShards = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "shards_max",
Help: "The maximum number of shards that the queue is allowed to run.",
ConstLabels: constLabels,
})
m.minNumShards = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "shards_min",
Help: "The minimum number of shards that the queue is allowed to run.",
ConstLabels: constLabels,
})
m.desiredNumShards = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "shards_desired",
Help: "The number of shards that the queues shard calculation wants to run based on the rate of samples in vs. samples out.",
ConstLabels: constLabels,
})
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.sentBytesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
Name: "bytes_total",
Help: "The total number of bytes of data (not metadata) sent by the queue after compression. Note that when exemplars over remote write is enabled the exemplars included in a remote write request count towards this metric.",
ConstLabels: constLabels,
})
m.metadataBytesTotal = prometheus.NewCounter(prometheus.CounterOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "metadata_bytes_total",
Help: "The total number of bytes of metadata sent by the queue after compression.",
ConstLabels: constLabels,
})
m.maxSamplesPerSend = prometheus.NewGauge(prometheus.GaugeOpts{
Namespace: namespace,
Subsystem: subsystem,
Name: "max_samples_per_send",
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
Help: "The maximum number of samples to be sent, in a single request, to the remote storage. Note that, when sending of exemplars over remote write is enabled, exemplars count towards this limt.",
ConstLabels: constLabels,
})
return m
}
func (m *queueManagerMetrics) register() {
if m.reg != nil {
m.reg.MustRegister(
m.samplesTotal,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.exemplarsTotal,
m.metadataTotal,
m.failedSamplesTotal,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.failedExemplarsTotal,
m.failedMetadataTotal,
m.retriedSamplesTotal,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.retriedExemplarsTotal,
m.retriedMetadataTotal,
m.droppedSamplesTotal,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.droppedExemplarsTotal,
m.enqueueRetriesTotal,
m.sentBatchDuration,
m.highestSentTimestamp,
m.pendingSamples,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.pendingExemplars,
m.shardCapacity,
m.numShards,
m.maxNumShards,
m.minNumShards,
m.desiredNumShards,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.sentBytesTotal,
m.metadataBytesTotal,
m.maxSamplesPerSend,
)
}
}
2017-05-10 02:44:13 -07:00
func (m *queueManagerMetrics) unregister() {
if m.reg != nil {
m.reg.Unregister(m.samplesTotal)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.reg.Unregister(m.exemplarsTotal)
m.reg.Unregister(m.metadataTotal)
m.reg.Unregister(m.failedSamplesTotal)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.reg.Unregister(m.failedExemplarsTotal)
m.reg.Unregister(m.failedMetadataTotal)
m.reg.Unregister(m.retriedSamplesTotal)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.reg.Unregister(m.retriedExemplarsTotal)
m.reg.Unregister(m.retriedMetadataTotal)
m.reg.Unregister(m.droppedSamplesTotal)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.reg.Unregister(m.droppedExemplarsTotal)
m.reg.Unregister(m.enqueueRetriesTotal)
m.reg.Unregister(m.sentBatchDuration)
m.reg.Unregister(m.highestSentTimestamp)
m.reg.Unregister(m.pendingSamples)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.reg.Unregister(m.pendingExemplars)
m.reg.Unregister(m.shardCapacity)
m.reg.Unregister(m.numShards)
m.reg.Unregister(m.maxNumShards)
m.reg.Unregister(m.minNumShards)
m.reg.Unregister(m.desiredNumShards)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
m.reg.Unregister(m.sentBytesTotal)
m.reg.Unregister(m.metadataBytesTotal)
m.reg.Unregister(m.maxSamplesPerSend)
}
}
// WriteClient defines an interface for sending a batch of samples to an
2017-05-10 02:44:13 -07:00
// external timeseries database.
type WriteClient interface {
2017-05-10 02:44:13 -07:00
// Store stores the given samples in the remote storage.
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
Store(context.Context, []byte) error
// Name uniquely identifies the remote storage.
2017-05-10 02:44:13 -07:00
Name() string
// Endpoint is the remote read or write endpoint for the storage client.
Endpoint() string
2017-05-10 02:44:13 -07:00
}
// QueueManager manages a queue of samples to be sent to the Storage
// indicated by the provided WriteClient. Implements writeTo interface
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// used by WAL Watcher.
2017-05-10 02:44:13 -07:00
type QueueManager struct {
lastSendTimestamp atomic.Int64
logger log.Logger
flushDeadline time.Duration
cfg config.QueueConfig
mcfg config.MetadataConfig
externalLabels labels.Labels
relabelConfigs []*relabel.Config
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
sendExemplars bool
watcher *wal.Watcher
metadataWatcher *MetadataWatcher
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
clientMtx sync.RWMutex
storeClient WriteClient
seriesMtx sync.Mutex // Covers seriesLabels and droppedSeries.
seriesLabels map[chunks.HeadSeriesRef]labels.Labels
droppedSeries map[chunks.HeadSeriesRef]struct{}
seriesSegmentMtx sync.Mutex // Covers seriesSegmentIndexes - if you also lock seriesMtx, take seriesMtx first.
seriesSegmentIndexes map[chunks.HeadSeriesRef]int
2017-05-10 02:44:13 -07:00
shards *shards
numShards int
reshardChan chan int
quit chan struct{}
wg sync.WaitGroup
2017-05-10 02:44:13 -07:00
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
dataIn, dataDropped, dataOut, dataOutDuration *ewmaRate
metrics *queueManagerMetrics
interner *pool
highestRecvTimestamp *maxTimestamp
2017-05-10 02:44:13 -07:00
}
// NewQueueManager builds a new QueueManager.
func NewQueueManager(
metrics *queueManagerMetrics,
watcherMetrics *wal.WatcherMetrics,
readerMetrics *wal.LiveReaderMetrics,
logger log.Logger,
walDir string,
samplesIn *ewmaRate,
cfg config.QueueConfig,
mCfg config.MetadataConfig,
externalLabels labels.Labels,
relabelConfigs []*relabel.Config,
client WriteClient,
flushDeadline time.Duration,
interner *pool,
highestRecvTimestamp *maxTimestamp,
sm ReadyScrapeManager,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
enableExemplarRemoteWrite bool,
) *QueueManager {
2017-08-11 11:45:52 -07:00
if logger == nil {
logger = log.NewNopLogger()
}
logger = log.With(logger, remoteName, client.Name(), endpoint, client.Endpoint())
2017-05-10 02:44:13 -07:00
t := &QueueManager{
logger: logger,
flushDeadline: flushDeadline,
2017-05-10 02:44:13 -07:00
cfg: cfg,
mcfg: mCfg,
2017-05-10 02:44:13 -07:00
externalLabels: externalLabels,
relabelConfigs: relabelConfigs,
storeClient: client,
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
sendExemplars: enableExemplarRemoteWrite,
2017-05-10 02:44:13 -07:00
seriesLabels: make(map[chunks.HeadSeriesRef]labels.Labels),
seriesSegmentIndexes: make(map[chunks.HeadSeriesRef]int),
droppedSeries: make(map[chunks.HeadSeriesRef]struct{}),
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
numShards: cfg.MinShards,
2017-05-10 02:44:13 -07:00
reshardChan: make(chan int),
quit: make(chan struct{}),
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
dataIn: samplesIn,
dataDropped: newEWMARate(ewmaWeight, shardUpdateDuration),
dataOut: newEWMARate(ewmaWeight, shardUpdateDuration),
dataOutDuration: newEWMARate(ewmaWeight, shardUpdateDuration),
metrics: metrics,
interner: interner,
highestRecvTimestamp: highestRecvTimestamp,
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
t.watcher = wal.NewWatcher(watcherMetrics, readerMetrics, logger, client.Name(), t, walDir, enableExemplarRemoteWrite)
if t.mcfg.Send {
t.metadataWatcher = NewMetadataWatcher(logger, sm, client.Name(), t, t.mcfg.SendInterval, flushDeadline)
}
t.shards = t.newShards()
2017-05-10 02:44:13 -07:00
return t
}
// AppendMetadata sends metadata the remote storage. Metadata is sent in batches, but is not parallelized.
func (t *QueueManager) AppendMetadata(ctx context.Context, metadata []scrape.MetricMetadata) {
mm := make([]prompb.MetricMetadata, 0, len(metadata))
for _, entry := range metadata {
mm = append(mm, prompb.MetricMetadata{
MetricFamilyName: entry.Metric,
Help: entry.Help,
Type: metricTypeToMetricTypeProto(entry.Type),
Unit: entry.Unit,
})
}
pBuf := proto.NewBuffer(nil)
numSends := int(math.Ceil(float64(len(metadata)) / float64(t.mcfg.MaxSamplesPerSend)))
for i := 0; i < numSends; i++ {
last := (i + 1) * t.mcfg.MaxSamplesPerSend
if last > len(metadata) {
last = len(metadata)
}
err := t.sendMetadataWithBackoff(ctx, mm[i*t.mcfg.MaxSamplesPerSend:last], pBuf)
if err != nil {
t.metrics.failedMetadataTotal.Add(float64(last - (i * t.mcfg.MaxSamplesPerSend)))
level.Error(t.logger).Log("msg", "non-recoverable error while sending metadata", "count", last-(i*t.mcfg.MaxSamplesPerSend), "err", err)
}
}
}
func (t *QueueManager) sendMetadataWithBackoff(ctx context.Context, metadata []prompb.MetricMetadata, pBuf *proto.Buffer) error {
// Build the WriteRequest with no samples.
req, _, err := buildWriteRequest(nil, metadata, pBuf, nil)
if err != nil {
return err
}
metadataCount := len(metadata)
attemptStore := func(try int) error {
span, ctx := opentracing.StartSpanFromContext(ctx, "Remote Metadata Send Batch")
defer span.Finish()
span.SetTag("metadata", metadataCount)
span.SetTag("try", try)
span.SetTag("remote_name", t.storeClient.Name())
span.SetTag("remote_url", t.storeClient.Endpoint())
begin := time.Now()
err := t.storeClient.Store(ctx, req)
t.metrics.sentBatchDuration.Observe(time.Since(begin).Seconds())
if err != nil {
span.LogKV("error", err)
ext.Error.Set(span, true)
return err
}
return nil
}
retry := func() {
t.metrics.retriedMetadataTotal.Add(float64(len(metadata)))
}
err = sendWriteRequestWithBackoff(ctx, t.cfg, t.logger, attemptStore, retry)
if err != nil {
return err
}
t.metrics.metadataTotal.Add(float64(len(metadata)))
t.metrics.metadataBytesTotal.Add(float64(len(req)))
return nil
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// Append queues a sample to be sent to the remote storage. Blocks until all samples are
// enqueued on their shards or a shutdown signal is received.
func (t *QueueManager) Append(samples []record.RefSample) bool {
outer:
for _, s := range samples {
t.seriesMtx.Lock()
lbls, ok := t.seriesLabels[s.Ref]
if !ok {
t.metrics.droppedSamplesTotal.Inc()
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
t.dataDropped.incr(1)
if _, ok := t.droppedSeries[s.Ref]; !ok {
level.Info(t.logger).Log("msg", "Dropped sample for series that was not explicitly dropped via relabelling", "ref", s.Ref)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
t.seriesMtx.Unlock()
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
continue
}
t.seriesMtx.Unlock()
// This will only loop if the queues are being resharded.
backoff := t.cfg.MinBackoff
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
for {
select {
case <-t.quit:
return false
default:
}
if t.shards.enqueue(s.Ref, sampleOrExemplar{
seriesLabels: lbls,
timestamp: s.T,
value: s.V,
isSample: true,
}) {
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
continue outer
}
t.metrics.enqueueRetriesTotal.Inc()
time.Sleep(time.Duration(backoff))
backoff = backoff * 2
if backoff > t.cfg.MaxBackoff {
backoff = t.cfg.MaxBackoff
}
}
}
return true
}
2017-05-10 02:44:13 -07:00
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
func (t *QueueManager) AppendExemplars(exemplars []record.RefExemplar) bool {
if !t.sendExemplars {
return true
}
outer:
for _, e := range exemplars {
t.seriesMtx.Lock()
lbls, ok := t.seriesLabels[e.Ref]
if !ok {
t.metrics.droppedExemplarsTotal.Inc()
// Track dropped exemplars in the same EWMA for sharding calc.
t.dataDropped.incr(1)
if _, ok := t.droppedSeries[e.Ref]; !ok {
level.Info(t.logger).Log("msg", "Dropped exemplar for series that was not explicitly dropped via relabelling", "ref", e.Ref)
}
t.seriesMtx.Unlock()
continue
}
t.seriesMtx.Unlock()
// This will only loop if the queues are being resharded.
backoff := t.cfg.MinBackoff
for {
select {
case <-t.quit:
return false
default:
}
if t.shards.enqueue(e.Ref, sampleOrExemplar{
seriesLabels: lbls,
timestamp: e.T,
value: e.V,
exemplarLabels: e.Labels,
}) {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
continue outer
}
t.metrics.enqueueRetriesTotal.Inc()
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
time.Sleep(time.Duration(backoff))
backoff = backoff * 2
if backoff > t.cfg.MaxBackoff {
backoff = t.cfg.MaxBackoff
}
2017-05-10 02:44:13 -07:00
}
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
return true
2017-05-10 02:44:13 -07:00
}
// Start the queue manager sending samples to the remote storage.
// Does not block.
func (t *QueueManager) Start() {
// Register and initialise some metrics.
t.metrics.register()
t.metrics.shardCapacity.Set(float64(t.cfg.Capacity))
t.metrics.maxNumShards.Set(float64(t.cfg.MaxShards))
t.metrics.minNumShards.Set(float64(t.cfg.MinShards))
t.metrics.desiredNumShards.Set(float64(t.cfg.MinShards))
t.metrics.maxSamplesPerSend.Set(float64(t.cfg.MaxSamplesPerSend))
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
t.shards.start(t.numShards)
t.watcher.Start()
if t.mcfg.Send {
t.metadataWatcher.Start()
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
2017-05-10 02:44:13 -07:00
t.wg.Add(2)
go t.updateShardsLoop()
go t.reshardLoop()
}
// Stop stops sending samples to the remote storage and waits for pending
// sends to complete.
func (t *QueueManager) Stop() {
2017-08-11 11:45:52 -07:00
level.Info(t.logger).Log("msg", "Stopping remote storage...")
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
defer level.Info(t.logger).Log("msg", "Remote storage stopped.")
2017-05-10 02:44:13 -07:00
close(t.quit)
t.wg.Wait()
// Wait for all QueueManager routines to end before stopping shards, metadata watcher, and WAL watcher. This
// is to ensure we don't end up executing a reshard and shards.stop() at the same time, which
// causes a closed channel panic.
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
t.shards.stop()
t.watcher.Stop()
if t.mcfg.Send {
t.metadataWatcher.Stop()
}
// On shutdown, release the strings in the labels from the intern pool.
t.seriesMtx.Lock()
for _, labels := range t.seriesLabels {
t.releaseLabels(labels)
}
t.seriesMtx.Unlock()
t.metrics.unregister()
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
// StoreSeries keeps track of which series we know about for lookups when sending samples to remote.
func (t *QueueManager) StoreSeries(series []record.RefSeries, index int) {
t.seriesMtx.Lock()
defer t.seriesMtx.Unlock()
t.seriesSegmentMtx.Lock()
defer t.seriesSegmentMtx.Unlock()
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
for _, s := range series {
// Just make sure all the Refs of Series will insert into seriesSegmentIndexes map for tracking.
t.seriesSegmentIndexes[s.Ref] = index
ls := processExternalLabels(s.Labels, t.externalLabels)
lbls := relabel.Process(ls, t.relabelConfigs...)
if len(lbls) == 0 {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
t.droppedSeries[s.Ref] = struct{}{}
continue
}
t.internLabels(lbls)
// We should not ever be replacing a series labels in the map, but just
// in case we do we need to ensure we do not leak the replaced interned
// strings.
if orig, ok := t.seriesLabels[s.Ref]; ok {
t.releaseLabels(orig)
}
t.seriesLabels[s.Ref] = lbls
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
}
2017-05-10 02:44:13 -07:00
// UpdateSeriesSegment updates the segment number held against the series,
// so we can trim older ones in SeriesReset.
func (t *QueueManager) UpdateSeriesSegment(series []record.RefSeries, index int) {
t.seriesSegmentMtx.Lock()
defer t.seriesSegmentMtx.Unlock()
for _, s := range series {
t.seriesSegmentIndexes[s.Ref] = index
}
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// SeriesReset is used when reading a checkpoint. WAL Watcher should have
// stored series records with the checkpoints index number, so we can now
// delete any ref ID's lower than that # from the two maps.
func (t *QueueManager) SeriesReset(index int) {
t.seriesMtx.Lock()
defer t.seriesMtx.Unlock()
t.seriesSegmentMtx.Lock()
defer t.seriesSegmentMtx.Unlock()
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// Check for series that are in segments older than the checkpoint
// that were not also present in the checkpoint.
for k, v := range t.seriesSegmentIndexes {
if v < index {
delete(t.seriesSegmentIndexes, k)
t.releaseLabels(t.seriesLabels[k])
delete(t.seriesLabels, k)
delete(t.droppedSeries, k)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
}
}
2017-08-11 11:45:52 -07:00
// SetClient updates the client used by a queue. Used when only client specific
// fields are updated to avoid restarting the queue.
func (t *QueueManager) SetClient(c WriteClient) {
t.clientMtx.Lock()
t.storeClient = c
t.clientMtx.Unlock()
}
func (t *QueueManager) client() WriteClient {
t.clientMtx.RLock()
defer t.clientMtx.RUnlock()
return t.storeClient
}
func (t *QueueManager) internLabels(lbls labels.Labels) {
for i, l := range lbls {
lbls[i].Name = t.interner.intern(l.Name)
lbls[i].Value = t.interner.intern(l.Value)
}
}
func (t *QueueManager) releaseLabels(ls labels.Labels) {
for _, l := range ls {
t.interner.release(l.Name)
t.interner.release(l.Value)
}
}
// processExternalLabels merges externalLabels into ls. If ls contains
// a label in externalLabels, the value in ls wins.
func processExternalLabels(ls, externalLabels labels.Labels) labels.Labels {
i, j, result := 0, 0, make(labels.Labels, 0, len(ls)+len(externalLabels))
for i < len(ls) && j < len(externalLabels) {
if ls[i].Name < externalLabels[j].Name {
result = append(result, labels.Label{
Name: ls[i].Name,
Value: ls[i].Value,
})
i++
} else if ls[i].Name > externalLabels[j].Name {
result = append(result, externalLabels[j])
j++
} else {
result = append(result, labels.Label{
Name: ls[i].Name,
Value: ls[i].Value,
})
i++
j++
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
}
return append(append(result, ls[i:]...), externalLabels[j:]...)
2017-05-10 02:44:13 -07:00
}
func (t *QueueManager) updateShardsLoop() {
defer t.wg.Done()
ticker := time.NewTicker(shardUpdateDuration)
defer ticker.Stop()
2017-05-10 02:44:13 -07:00
for {
select {
case <-ticker.C:
desiredShards := t.calculateDesiredShards()
if !t.shouldReshard(desiredShards) {
continue
}
// Resharding can take some time, and we want this loop
// to stay close to shardUpdateDuration.
select {
case t.reshardChan <- desiredShards:
level.Info(t.logger).Log("msg", "Remote storage resharding", "from", t.numShards, "to", desiredShards)
t.numShards = desiredShards
default:
level.Info(t.logger).Log("msg", "Currently resharding, skipping.")
}
2017-05-10 02:44:13 -07:00
case <-t.quit:
return
}
}
}
// shouldReshard returns if resharding should occur
func (t *QueueManager) shouldReshard(desiredShards int) bool {
if desiredShards == t.numShards {
return false
}
// We shouldn't reshard if Prometheus hasn't been able to send to the
// remote endpoint successfully within some period of time.
minSendTimestamp := time.Now().Add(-2 * time.Duration(t.cfg.BatchSendDeadline)).Unix()
lsts := t.lastSendTimestamp.Load()
if lsts < minSendTimestamp {
level.Warn(t.logger).Log("msg", "Skipping resharding, last successful send was beyond threshold", "lastSendTimestamp", lsts, "minSendTimestamp", minSendTimestamp)
return false
}
return true
}
// calculateDesiredShards returns the number of desired shards, which will be
// the current QueueManager.numShards if resharding should not occur for reasons
// outlined in this functions implementation. It is up to the caller to reshard, or not,
// based on the return value.
func (t *QueueManager) calculateDesiredShards() int {
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
t.dataOut.tick()
t.dataDropped.tick()
t.dataOutDuration.tick()
2017-05-10 02:44:13 -07:00
// We use the number of incoming samples as a prediction of how much work we
// will need to do next iteration. We add to this any pending samples
// (received - send) so we can catch up with any backlog. We use the average
// outgoing batch latency to work out how many shards we need.
var (
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
dataInRate = t.dataIn.rate()
dataOutRate = t.dataOut.rate()
dataKeptRatio = dataOutRate / (t.dataDropped.rate() + dataOutRate)
dataOutDuration = t.dataOutDuration.rate() / float64(time.Second)
dataPendingRate = dataInRate*dataKeptRatio - dataOutRate
highestSent = t.metrics.highestSentTimestamp.Get()
highestRecv = t.highestRecvTimestamp.Get()
delay = highestRecv - highestSent
dataPending = delay * dataInRate * dataKeptRatio
2017-05-10 02:44:13 -07:00
)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
if dataOutRate <= 0 {
return t.numShards
2017-05-10 02:44:13 -07:00
}
// When behind we will try to catch up on a proporation of samples per tick.
// This works similarly to an integral accumulator in that pending samples
// is the result of the error integral.
const integralGain = 0.1 / float64(shardUpdateDuration/time.Second)
2017-05-10 02:44:13 -07:00
var (
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
timePerSample = dataOutDuration / dataOutRate
desiredShards = timePerSample * (dataInRate*dataKeptRatio + integralGain*dataPending)
2017-05-10 02:44:13 -07:00
)
t.metrics.desiredNumShards.Set(desiredShards)
level.Debug(t.logger).Log("msg", "QueueManager.calculateDesiredShards",
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
"dataInRate", dataInRate,
"dataOutRate", dataOutRate,
"dataKeptRatio", dataKeptRatio,
"dataPendingRate", dataPendingRate,
"dataPending", dataPending,
"dataOutDuration", dataOutDuration,
"timePerSample", timePerSample,
"desiredShards", desiredShards,
"highestSent", highestSent,
"highestRecv", highestRecv,
)
2017-05-10 02:44:13 -07:00
// Changes in the number of shards must be greater than shardToleranceFraction.
var (
lowerBound = float64(t.numShards) * (1. - shardToleranceFraction)
upperBound = float64(t.numShards) * (1. + shardToleranceFraction)
)
2017-08-11 11:45:52 -07:00
level.Debug(t.logger).Log("msg", "QueueManager.updateShardsLoop",
"lowerBound", lowerBound, "desiredShards", desiredShards, "upperBound", upperBound)
2017-05-10 02:44:13 -07:00
if lowerBound <= desiredShards && desiredShards <= upperBound {
return t.numShards
2017-05-10 02:44:13 -07:00
}
numShards := int(math.Ceil(desiredShards))
// Do not downshard if we are more than ten seconds back.
if numShards < t.numShards && delay > 10.0 {
level.Debug(t.logger).Log("msg", "Not downsharding due to being too far behind")
return t.numShards
}
2017-05-10 02:44:13 -07:00
if numShards > t.cfg.MaxShards {
numShards = t.cfg.MaxShards
} else if numShards < t.cfg.MinShards {
numShards = t.cfg.MinShards
2017-05-10 02:44:13 -07:00
}
return numShards
2017-05-10 02:44:13 -07:00
}
func (t *QueueManager) reshardLoop() {
defer t.wg.Done()
for {
select {
case numShards := <-t.reshardChan:
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// We start the newShards after we have stopped (the therefore completely
// flushed) the oldShards, to guarantee we only every deliver samples in
// order.
t.shards.stop()
t.shards.start(numShards)
2017-05-10 02:44:13 -07:00
case <-t.quit:
return
}
}
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
func (t *QueueManager) newShards() *shards {
s := &shards{
qm: t,
done: make(chan struct{}),
}
return s
}
2017-05-10 02:44:13 -07:00
type shards struct {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
mtx sync.RWMutex // With the WAL, this is never actually contended.
qm *QueueManager
queues []*queue
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
// So we can accurately track how many of each are lost during shard shutdowns.
enqueuedSamples atomic.Int64
enqueuedExemplars atomic.Int64
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// Emulate a wait group with a channel and an atomic int, as you
// cannot select on a wait group.
done chan struct{}
running atomic.Int32
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// Soft shutdown context will prevent new enqueues and deadlocks.
softShutdown chan struct{}
// Hard shutdown context is used to terminate outgoing HTTP connections
// after giving them a chance to terminate.
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
hardShutdown context.CancelFunc
samplesDroppedOnHardShutdown atomic.Uint32
exemplarsDroppedOnHardShutdown atomic.Uint32
2017-05-10 02:44:13 -07:00
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// start the shards; must be called before any call to enqueue.
func (s *shards) start(n int) {
s.mtx.Lock()
defer s.mtx.Unlock()
s.qm.metrics.pendingSamples.Set(0)
s.qm.metrics.numShards.Set(float64(n))
newQueues := make([]*queue, n)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
for i := 0; i < n; i++ {
newQueues[i] = newQueue(s.qm.cfg.MaxSamplesPerSend, s.qm.cfg.Capacity)
2017-05-10 02:44:13 -07:00
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
s.queues = newQueues
var hardShutdownCtx context.Context
hardShutdownCtx, s.hardShutdown = context.WithCancel(context.Background())
s.softShutdown = make(chan struct{})
s.running.Store(int32(n))
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
s.done = make(chan struct{})
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.samplesDroppedOnHardShutdown.Store(0)
s.exemplarsDroppedOnHardShutdown.Store(0)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
for i := 0; i < n; i++ {
go s.runShard(hardShutdownCtx, i, newQueues[i])
2017-05-10 02:44:13 -07:00
}
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// stop the shards; subsequent call to enqueue will return false.
func (s *shards) stop() {
// Attempt a clean shutdown, but only wait flushDeadline for all the shards
// to cleanly exit. As we're doing RPCs, enqueue can block indefinitely.
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// We must be able so call stop concurrently, hence we can only take the
// RLock here.
s.mtx.RLock()
close(s.softShutdown)
s.mtx.RUnlock()
// Enqueue should now be unblocked, so we can take the write lock. This
// also ensures we don't race with writes to the queues, and get a panic:
// send on closed channel.
s.mtx.Lock()
defer s.mtx.Unlock()
for _, queue := range s.queues {
go queue.FlushAndShutdown(s.done)
2017-05-10 02:44:13 -07:00
}
select {
case <-s.done:
return
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
case <-time.After(s.qm.flushDeadline):
}
// Force an unclean shutdown.
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
s.hardShutdown()
<-s.done
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
if dropped := s.samplesDroppedOnHardShutdown.Load(); dropped > 0 {
level.Error(s.qm.logger).Log("msg", "Failed to flush all samples on shutdown", "count", dropped)
}
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
if dropped := s.exemplarsDroppedOnHardShutdown.Load(); dropped > 0 {
level.Error(s.qm.logger).Log("msg", "Failed to flush all exemplars on shutdown", "count", dropped)
}
2017-05-10 02:44:13 -07:00
}
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
// enqueue data (sample or exemplar). If we are currently in the process of shutting down or resharding,
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
// will return false; in this case, you should back off and retry.
func (s *shards) enqueue(ref chunks.HeadSeriesRef, data sampleOrExemplar) bool {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
s.mtx.RLock()
defer s.mtx.RUnlock()
2017-05-10 02:44:13 -07:00
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
select {
case <-s.softShutdown:
return false
default:
}
2017-05-10 02:44:13 -07:00
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
shard := uint64(ref) % uint64(len(s.queues))
2017-05-10 02:44:13 -07:00
select {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
case <-s.softShutdown:
return false
default:
appended := s.queues[shard].Append(data, s.softShutdown)
if !appended {
return false
}
if data.isSample {
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.pendingSamples.Inc()
s.enqueuedSamples.Inc()
} else {
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.pendingExemplars.Inc()
s.enqueuedExemplars.Inc()
}
2017-05-10 02:44:13 -07:00
return true
}
}
type queue struct {
batch []sampleOrExemplar
batchQueue chan []sampleOrExemplar
// Since we know there are a limited number of batches out, using a stack
// is easy and safe so a sync.Pool is not necessary.
batchPool [][]sampleOrExemplar
// This mutex covers adding and removing batches from the batchPool.
poolMux sync.Mutex
}
type sampleOrExemplar struct {
seriesLabels labels.Labels
value float64
timestamp int64
exemplarLabels labels.Labels
isSample bool
}
func newQueue(batchSize, capacity int) *queue {
batches := capacity / batchSize
return &queue{
batch: make([]sampleOrExemplar, 0, batchSize),
batchQueue: make(chan []sampleOrExemplar, batches),
// batchPool should have capacity for everything in the channel + 1 for
// the batch being processed.
batchPool: make([][]sampleOrExemplar, 0, batches+1),
}
}
func (q *queue) Append(datum sampleOrExemplar, stop <-chan struct{}) bool {
q.batch = append(q.batch, datum)
if len(q.batch) == cap(q.batch) {
select {
case q.batchQueue <- q.batch:
q.batch = q.newBatch(cap(q.batch))
return true
case <-stop:
// Remove the sample we just appended. It will get retried.
q.batch = q.batch[:len(q.batch)-1]
return false
}
}
return true
}
func (q *queue) Chan() <-chan []sampleOrExemplar {
return q.batchQueue
}
// Batch returns the current batch and allocates a new batch. Must not be
// called concurrently with Append.
func (q *queue) Batch() []sampleOrExemplar {
batch := q.batch
q.batch = q.newBatch(cap(batch))
return batch
}
// ReturnForReuse adds the batch buffer back to the internal pool.
func (q *queue) ReturnForReuse(batch []sampleOrExemplar) {
q.poolMux.Lock()
defer q.poolMux.Unlock()
if len(q.batchPool) < cap(q.batchPool) {
q.batchPool = append(q.batchPool, batch[:0])
}
}
// FlushAndShutdown stops the queue and flushes any samples. No appends can be
// made after this is called.
func (q *queue) FlushAndShutdown(done <-chan struct{}) {
if len(q.batch) > 0 {
select {
case q.batchQueue <- q.batch:
case <-done:
// The shard has been hard shut down, so no more samples can be
// sent. Drop everything left in the queue.
}
}
q.batch = nil
close(q.batchQueue)
}
func (q *queue) newBatch(capacity int) []sampleOrExemplar {
q.poolMux.Lock()
defer q.poolMux.Unlock()
batches := len(q.batchPool)
if batches > 0 {
batch := q.batchPool[batches-1]
q.batchPool = q.batchPool[:batches-1]
return batch
}
return make([]sampleOrExemplar, 0, capacity)
}
func (s *shards) runShard(ctx context.Context, shardID int, queue *queue) {
defer func() {
if s.running.Dec() == 0 {
close(s.done)
}
}()
shardNum := strconv.Itoa(shardID)
2017-05-10 02:44:13 -07:00
// Send batches of at most MaxSamplesPerSend samples to the remote storage.
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
// If we have fewer samples than that, flush them out after a deadline anyways.
var (
max = s.qm.cfg.MaxSamplesPerSend
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
pBuf = proto.NewBuffer(nil)
buf []byte
)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
if s.qm.sendExemplars {
max += int(float64(max) * 0.1)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
}
batchQueue := queue.Chan()
pendingData := make([]prompb.TimeSeries, max)
for i := range pendingData {
pendingData[i].Samples = []prompb.Sample{{}}
if s.qm.sendExemplars {
pendingData[i].Exemplars = []prompb.Exemplar{{}}
}
}
timer := time.NewTimer(time.Duration(s.qm.cfg.BatchSendDeadline))
2018-03-12 07:27:48 -07:00
stop := func() {
if !timer.Stop() {
select {
case <-timer.C:
default:
}
}
2018-03-12 07:27:48 -07:00
}
defer stop()
2017-05-10 02:44:13 -07:00
for {
select {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
case <-ctx.Done():
// In this case we drop all samples in the buffer and the queue.
// Remove them from pending and mark them as failed.
droppedSamples := int(s.enqueuedSamples.Load())
droppedExemplars := int(s.enqueuedExemplars.Load())
s.qm.metrics.pendingSamples.Sub(float64(droppedSamples))
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.pendingExemplars.Sub(float64(droppedExemplars))
s.qm.metrics.failedSamplesTotal.Add(float64(droppedSamples))
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.failedExemplarsTotal.Add(float64(droppedExemplars))
s.samplesDroppedOnHardShutdown.Add(uint32(droppedSamples))
s.exemplarsDroppedOnHardShutdown.Add(uint32(droppedExemplars))
return
case batch, ok := <-batchQueue:
2017-05-10 02:44:13 -07:00
if !ok {
return
}
nPendingSamples, nPendingExemplars := s.populateTimeSeries(batch, pendingData)
queue.ReturnForReuse(batch)
n := nPendingSamples + nPendingExemplars
s.sendSamples(ctx, pendingData[:n], nPendingSamples, nPendingExemplars, pBuf, &buf)
stop()
timer.Reset(time.Duration(s.qm.cfg.BatchSendDeadline))
case <-timer.C:
// We need to take the lock when getting a batch to avoid
// concurrent Appends. Generally this will only happen on low
// traffic instances.
s.mtx.Lock()
// First, we need to see if we can happen to get a batch from the queue if it filled while acquiring the lock.
var batch []sampleOrExemplar
select {
case batch = <-batchQueue:
default:
batch = queue.Batch()
}
s.mtx.Unlock()
if len(batch) > 0 {
nPendingSamples, nPendingExemplars := s.populateTimeSeries(batch, pendingData)
n := nPendingSamples + nPendingExemplars
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
level.Debug(s.qm.logger).Log("msg", "runShard timer ticked, sending buffered data", "samples", nPendingSamples, "exemplars", nPendingExemplars, "shard", shardNum)
s.sendSamples(ctx, pendingData[:n], nPendingSamples, nPendingExemplars, pBuf, &buf)
2017-05-10 02:44:13 -07:00
}
queue.ReturnForReuse(batch)
timer.Reset(time.Duration(s.qm.cfg.BatchSendDeadline))
2017-05-10 02:44:13 -07:00
}
}
}
func (s *shards) populateTimeSeries(batch []sampleOrExemplar, pendingData []prompb.TimeSeries) (int, int) {
var nPendingSamples, nPendingExemplars int
for nPending, d := range batch {
pendingData[nPending].Samples = pendingData[nPending].Samples[:0]
if s.qm.sendExemplars {
pendingData[nPending].Exemplars = pendingData[nPending].Exemplars[:0]
}
// Number of pending samples is limited by the fact that sendSamples (via sendSamplesWithBackoff)
// retries endlessly, so once we reach max samples, if we can never send to the endpoint we'll
// stop reading from the queue. This makes it safe to reference pendingSamples by index.
if d.isSample {
pendingData[nPending].Labels = labelsToLabelsProto(d.seriesLabels, pendingData[nPending].Labels)
pendingData[nPending].Samples = append(pendingData[nPending].Samples, prompb.Sample{
Value: d.value,
Timestamp: d.timestamp,
})
nPendingSamples++
} else {
pendingData[nPending].Labels = labelsToLabelsProto(d.seriesLabels, pendingData[nPending].Labels)
pendingData[nPending].Exemplars = append(pendingData[nPending].Exemplars, prompb.Exemplar{
Labels: labelsToLabelsProto(d.exemplarLabels, nil),
Value: d.value,
Timestamp: d.timestamp,
})
nPendingExemplars++
}
}
return nPendingSamples, nPendingExemplars
}
func (s *shards) sendSamples(ctx context.Context, samples []prompb.TimeSeries, sampleCount, exemplarCount int, pBuf *proto.Buffer, buf *[]byte) {
2017-05-10 02:44:13 -07:00
begin := time.Now()
err := s.sendSamplesWithBackoff(ctx, samples, sampleCount, exemplarCount, pBuf, buf)
if err != nil {
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
level.Error(s.qm.logger).Log("msg", "non-recoverable error", "count", sampleCount, "exemplarCount", exemplarCount, "err", err)
s.qm.metrics.failedSamplesTotal.Add(float64(sampleCount))
s.qm.metrics.failedExemplarsTotal.Add(float64(exemplarCount))
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
2017-05-10 02:44:13 -07:00
2018-04-08 02:51:54 -07:00
// These counters are used to calculate the dynamic sharding, and as such
2017-05-10 02:44:13 -07:00
// should be maintained irrespective of success or failure.
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.dataOut.incr(int64(len(samples)))
s.qm.dataOutDuration.incr(int64(time.Since(begin)))
s.qm.lastSendTimestamp.Store(time.Now().Unix())
// Pending samples/exemplars also should be subtracted as an error means
// they will not be retried.
s.qm.metrics.pendingSamples.Sub(float64(sampleCount))
s.qm.metrics.pendingExemplars.Sub(float64(exemplarCount))
s.enqueuedSamples.Sub(int64(sampleCount))
s.enqueuedExemplars.Sub(int64(exemplarCount))
2017-05-10 02:44:13 -07:00
}
// sendSamples to the remote storage with backoff for recoverable errors.
func (s *shards) sendSamplesWithBackoff(ctx context.Context, samples []prompb.TimeSeries, sampleCount, exemplarCount int, pBuf *proto.Buffer, buf *[]byte) error {
// Build the WriteRequest with no metadata.
req, highest, err := buildWriteRequest(samples, nil, pBuf, *buf)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
if err != nil {
// Failing to build the write request is non-recoverable, since it will
// only error if marshaling the proto to bytes fails.
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
return err
}
reqSize := len(req)
*buf = req
// An anonymous function allows us to defer the completion of our per-try spans
// without causing a memory leak, and it has the nice effect of not propagating any
// parameters for sendSamplesWithBackoff/3.
attemptStore := func(try int) error {
span, ctx := opentracing.StartSpanFromContext(ctx, "Remote Send Batch")
defer span.Finish()
span.SetTag("samples", sampleCount)
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
if exemplarCount > 0 {
span.SetTag("exemplars", exemplarCount)
}
span.SetTag("request_size", reqSize)
span.SetTag("try", try)
span.SetTag("remote_name", s.qm.storeClient.Name())
span.SetTag("remote_url", s.qm.storeClient.Endpoint())
begin := time.Now()
s.qm.metrics.samplesTotal.Add(float64(sampleCount))
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.exemplarsTotal.Add(float64(exemplarCount))
err := s.qm.client().Store(ctx, *buf)
s.qm.metrics.sentBatchDuration.Observe(time.Since(begin).Seconds())
if err != nil {
span.LogKV("error", err)
ext.Error.Set(span, true)
return err
}
return nil
}
onRetry := func() {
s.qm.metrics.retriedSamplesTotal.Add(float64(sampleCount))
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.retriedExemplarsTotal.Add(float64(exemplarCount))
}
err = sendWriteRequestWithBackoff(ctx, s.qm.cfg, s.qm.logger, attemptStore, onRetry)
if errors.Is(err, context.Canceled) {
// When there is resharding, we cancel the context for this queue, which means the data is not sent.
// So we exit early to not update the metrics.
return err
}
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
s.qm.metrics.sentBytesTotal.Add(float64(reqSize))
s.qm.metrics.highestSentTimestamp.Set(float64(highest / 1000))
return err
}
func sendWriteRequestWithBackoff(ctx context.Context, cfg config.QueueConfig, l log.Logger, attempt func(int) error, onRetry func()) error {
backoff := cfg.MinBackoff
sleepDuration := model.Duration(0)
try := 0
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
2017-05-10 02:44:13 -07:00
err := attempt(try)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
if err == nil {
return nil
}
2017-05-10 02:44:13 -07:00
// If the error is unrecoverable, we should not retry.
backoffErr, ok := err.(RecoverableError)
if !ok {
return err
}
sleepDuration = backoff
if backoffErr.retryAfter > 0 {
sleepDuration = backoffErr.retryAfter
level.Info(l).Log("msg", "Retrying after duration specified by Retry-After header", "duration", sleepDuration)
} else if backoffErr.retryAfter < 0 {
level.Debug(l).Log("msg", "retry-after cannot be in past, retrying using default backoff mechanism")
}
select {
case <-ctx.Done():
case <-time.After(time.Duration(sleepDuration)):
}
// If we make it this far, we've encountered a recoverable error and will retry.
onRetry()
level.Warn(l).Log("msg", "Failed to send batch, retrying", "err", err)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
backoff = sleepDuration * 2
if backoff > cfg.MaxBackoff {
backoff = cfg.MaxBackoff
2017-05-10 02:44:13 -07:00
}
try++
2017-05-10 02:44:13 -07:00
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
func buildWriteRequest(samples []prompb.TimeSeries, metadata []prompb.MetricMetadata, pBuf *proto.Buffer, buf []byte) ([]byte, int64, error) {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
var highest int64
for _, ts := range samples {
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
// At the moment we only ever append a TimeSeries with a single sample or exemplar in it.
if len(ts.Samples) > 0 && ts.Samples[0].Timestamp > highest {
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
highest = ts.Samples[0].Timestamp
}
Add Exemplar Remote Write support (#8296) * Write exemplars to the WAL and send them over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Update example for exemplars, print data in a more obvious format. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add metrics for remote write of exemplars. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Fix incorrect slices passed to send in remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * We need to unregister the new metrics. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> * Order of exemplar append vs write exemplar to WAL needs to change. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Several fixes to prevent sending uninitialized or incorrect samples with an exemplar. Fix dropping exemplar for missing series. Add tests for queue_manager sending exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Store both samples and exemplars in the same timeseries buffer to remove the alloc when building final request, keep sub-slices in separate buffers for re-use Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Condense sample/exemplar delivery tests to parameterized sub-tests Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename test methods for clarity now that they also handle exemplars Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Rename counter variable. Fix instances where metrics were not updated correctly Signed-off-by: Martin Disibio <mdisibio@gmail.com> * Add exemplars to LoadWAL benchmark Signed-off-by: Callum Styan <callumstyan@gmail.com> * last exemplars timestamp metric needs to convert value to seconds with ms precision Signed-off-by: Callum Styan <callumstyan@gmail.com> * Process exemplar records in a separate go routine when loading the WAL. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address review comments related to clarifying comments and variable names. Also refactor sample/exemplar to enqueue prompb types. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Regenerate types proto with comments, update protoc version again. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Put remote write of exemplars behind a feature flag. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address some of Ganesh's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Move exemplar remote write feature flag to a config file field. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address Bartek's review comments. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Don't allocate exemplar buffers in queue_manager if we're not going to send exemplars over remote write. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add ValidateExemplar function, validate exemplars when appending to head and log them all to WAL before adding them to exemplar storage. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address more reivew comments from Ganesh. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Add exemplar total label length check. Signed-off-by: Callum Styan <callumstyan@gmail.com> * Address a few last review comments Signed-off-by: Callum Styan <callumstyan@gmail.com> Co-authored-by: Martin Disibio <mdisibio@gmail.com>
2021-05-06 13:53:52 -07:00
if len(ts.Exemplars) > 0 && ts.Exemplars[0].Timestamp > highest {
highest = ts.Exemplars[0].Timestamp
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
req := &prompb.WriteRequest{
Timeseries: samples,
Metadata: metadata,
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
}
if pBuf == nil {
pBuf = proto.NewBuffer(nil) // For convenience in tests. Not efficient.
} else {
pBuf.Reset()
}
err := pBuf.Marshal(req)
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
if err != nil {
return nil, highest, err
}
2017-05-10 02:44:13 -07:00
// snappy uses len() to see if it needs to allocate a new slice. Make the
// buffer as long as possible.
if buf != nil {
buf = buf[0:cap(buf)]
}
compressed := snappy.Encode(buf, pBuf.Bytes())
Tail the TSDB WAL for remote_write This change switches the remote_write API to use the TSDB WAL. This should reduce memory usage and prevent sample loss when the remote end point is down. We use the new LiveReader from TSDB to tail WAL segments. Logic for finding the tracking segment is included in this PR. The WAL is tailed once for each remote_write endpoint specified. Reading from the segment is based on a ticker rather than relying on fsnotify write events, which were found to be complicated and unreliable in early prototypes. Enqueuing a sample for sending via remote_write can now block, to provide back pressure. Queues are still required to acheive parallelism and batching. We have updated the queue config based on new defaults for queue capacity and pending samples values - much smaller values are now possible. The remote_write resharding code has been updated to prevent deadlocks, and extra tests have been added for these cases. As part of this change, we attempt to guarantee that samples are not lost; however this initial version doesn't guarantee this across Prometheus restarts or non-retryable errors from the remote end (eg 400s). This changes also includes the following optimisations: - only marshal the proto request once, not once per retry - maintain a single copy of the labels for given series to reduce GC pressure Other minor tweaks: - only reshard if we've also successfully sent recently - add pending samples, latest sent timestamp, WAL events processed metrics Co-authored-by: Chris Marchbanks <csmarchbanks.com> (initial prototype) Co-authored-by: Tom Wilkie <tom.wilkie@gmail.com> (sharding changes) Signed-off-by: Callum Styan <callumstyan@gmail.com>
2018-09-07 14:26:04 -07:00
return compressed, highest, nil
2017-05-10 02:44:13 -07:00
}