2015-01-21 11:07:45 -08:00
|
|
|
// Copyright 2013 The Prometheus Authors
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
// 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 (
|
|
|
|
"time"
|
|
|
|
|
2015-05-20 09:10:29 -07:00
|
|
|
"github.com/prometheus/client_golang/prometheus"
|
2015-10-03 01:21:43 -07:00
|
|
|
"github.com/prometheus/common/log"
|
2015-08-20 08:18:46 -07:00
|
|
|
"github.com/prometheus/common/model"
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
)
|
|
|
|
|
|
|
|
const (
|
2015-03-29 17:42:04 -07:00
|
|
|
// The maximum number of concurrent send requests to the remote storage.
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
maxConcurrentSends = 10
|
2015-03-29 17:42:04 -07:00
|
|
|
// The maximum number of samples to fit into a single request to the remote storage.
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
maxSamplesPerSend = 100
|
|
|
|
// The deadline after which to send queued samples even if the maximum batch
|
|
|
|
// size has not been reached.
|
|
|
|
batchSendDeadline = 5 * time.Second
|
|
|
|
)
|
|
|
|
|
2014-06-18 10:43:15 -07:00
|
|
|
// String constants for instrumentation.
|
|
|
|
const (
|
2014-07-23 10:55:33 -07:00
|
|
|
namespace = "prometheus"
|
2014-10-22 10:21:23 -07:00
|
|
|
subsystem = "remote_storage"
|
2014-07-23 10:55:33 -07:00
|
|
|
|
2014-06-18 10:43:15 -07:00
|
|
|
result = "result"
|
|
|
|
success = "success"
|
|
|
|
failure = "failure"
|
|
|
|
dropped = "dropped"
|
|
|
|
)
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// StorageClient defines an interface for sending a batch of samples to an
|
|
|
|
// external timeseries database.
|
|
|
|
type StorageClient interface {
|
2015-04-02 11:20:00 -07:00
|
|
|
// Store stores the given samples in the remote storage.
|
2015-08-20 08:18:46 -07:00
|
|
|
Store(model.Samples) error
|
2015-04-02 11:20:00 -07:00
|
|
|
// Name identifies the remote storage implementation.
|
|
|
|
Name() string
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// StorageQueueManager manages a queue of samples to be sent to the Storage
|
|
|
|
// indicated by the provided StorageClient.
|
|
|
|
type StorageQueueManager struct {
|
|
|
|
tsdb StorageClient
|
2015-08-20 08:18:46 -07:00
|
|
|
queue chan *model.Sample
|
|
|
|
pendingSamples model.Samples
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
sendSemaphore chan bool
|
|
|
|
drained chan bool
|
2014-06-18 10:43:15 -07:00
|
|
|
|
2014-07-23 10:55:33 -07:00
|
|
|
samplesCount *prometheus.CounterVec
|
2015-01-13 09:33:35 -08:00
|
|
|
sendLatency prometheus.Summary
|
2015-09-16 07:22:18 -07:00
|
|
|
failedBatches prometheus.Counter
|
|
|
|
failedSamples prometheus.Counter
|
2014-07-23 10:55:33 -07:00
|
|
|
queueLength prometheus.Gauge
|
|
|
|
queueCapacity prometheus.Metric
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// NewStorageQueueManager builds a new StorageQueueManager.
|
|
|
|
func NewStorageQueueManager(tsdb StorageClient, queueCapacity int) *StorageQueueManager {
|
2015-04-02 11:20:00 -07:00
|
|
|
constLabels := prometheus.Labels{
|
|
|
|
"type": tsdb.Name(),
|
|
|
|
}
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
return &StorageQueueManager{
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
tsdb: tsdb,
|
2015-08-20 08:18:46 -07:00
|
|
|
queue: make(chan *model.Sample, queueCapacity),
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
sendSemaphore: make(chan bool, maxConcurrentSends),
|
|
|
|
drained: make(chan bool),
|
2014-06-18 10:43:15 -07:00
|
|
|
|
|
|
|
samplesCount: prometheus.NewCounterVec(
|
|
|
|
prometheus.CounterOpts{
|
2015-04-02 11:20:00 -07:00
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "sent_samples_total",
|
|
|
|
Help: "Total number of processed samples to be sent to remote storage.",
|
|
|
|
ConstLabels: constLabels,
|
2014-06-18 10:43:15 -07:00
|
|
|
},
|
|
|
|
[]string{result},
|
|
|
|
),
|
2015-01-13 09:33:35 -08:00
|
|
|
sendLatency: prometheus.NewSummary(prometheus.SummaryOpts{
|
2015-04-02 11:20:00 -07:00
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
2015-09-16 07:22:18 -07:00
|
|
|
Name: "send_latency_seconds",
|
2015-04-02 11:20:00 -07:00
|
|
|
Help: "Latency quantiles for sending sample batches to the remote storage.",
|
|
|
|
ConstLabels: constLabels,
|
2015-01-13 09:33:35 -08:00
|
|
|
}),
|
2015-09-16 07:22:18 -07:00
|
|
|
failedBatches: prometheus.NewCounter(prometheus.CounterOpts{
|
2015-04-02 11:20:00 -07:00
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
2015-09-16 07:22:18 -07:00
|
|
|
Name: "failed_batches_total",
|
|
|
|
Help: "Total number of sample batches that encountered an error while being sent to the remote storage.",
|
|
|
|
ConstLabels: constLabels,
|
|
|
|
}),
|
|
|
|
failedSamples: prometheus.NewCounter(prometheus.CounterOpts{
|
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "failed_samples_total",
|
|
|
|
Help: "Total number of samples that encountered an error while being sent to the remote storage.",
|
2015-04-02 11:20:00 -07:00
|
|
|
ConstLabels: constLabels,
|
2015-01-13 09:33:35 -08:00
|
|
|
}),
|
2014-07-23 10:55:33 -07:00
|
|
|
queueLength: prometheus.NewGauge(prometheus.GaugeOpts{
|
2015-04-02 11:20:00 -07:00
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "queue_length",
|
|
|
|
Help: "The number of processed samples queued to be sent to the remote storage.",
|
|
|
|
ConstLabels: constLabels,
|
2014-07-23 10:55:33 -07:00
|
|
|
}),
|
|
|
|
queueCapacity: prometheus.MustNewConstMetric(
|
|
|
|
prometheus.NewDesc(
|
|
|
|
prometheus.BuildFQName(namespace, subsystem, "queue_capacity"),
|
2015-03-29 17:42:04 -07:00
|
|
|
"The capacity of the queue of samples to be sent to the remote storage.",
|
2015-04-02 11:20:00 -07:00
|
|
|
nil,
|
|
|
|
constLabels,
|
2014-07-23 10:55:33 -07:00
|
|
|
),
|
|
|
|
prometheus.GaugeValue,
|
|
|
|
float64(queueCapacity),
|
2014-06-18 10:43:15 -07:00
|
|
|
),
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// Append queues a sample to be sent to the remote storage. It drops the
|
2016-01-27 10:07:46 -08:00
|
|
|
// sample on the floor if the queue is full.
|
2016-02-02 05:01:44 -08:00
|
|
|
// Always returns nil.
|
|
|
|
func (t *StorageQueueManager) Append(s *model.Sample) error {
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
select {
|
|
|
|
case t.queue <- s:
|
|
|
|
default:
|
2015-03-14 19:36:15 -07:00
|
|
|
t.samplesCount.WithLabelValues(dropped).Inc()
|
2015-05-20 09:10:29 -07:00
|
|
|
log.Warn("Remote storage queue full, discarding sample.")
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
2016-02-02 05:01:44 -08:00
|
|
|
return nil
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// Stop stops sending samples to the remote storage and waits for pending
|
|
|
|
// sends to complete.
|
|
|
|
func (t *StorageQueueManager) Stop() {
|
2015-05-20 09:10:29 -07:00
|
|
|
log.Infof("Stopping remote storage...")
|
2014-06-18 10:43:15 -07:00
|
|
|
close(t.queue)
|
|
|
|
<-t.drained
|
|
|
|
for i := 0; i < maxConcurrentSends; i++ {
|
|
|
|
t.sendSemaphore <- true
|
|
|
|
}
|
2015-05-20 09:10:29 -07:00
|
|
|
log.Info("Remote storage stopped.")
|
2014-06-18 10:43:15 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
// Describe implements prometheus.Collector.
|
2015-03-29 17:42:04 -07:00
|
|
|
func (t *StorageQueueManager) Describe(ch chan<- *prometheus.Desc) {
|
2014-06-18 10:43:15 -07:00
|
|
|
t.samplesCount.Describe(ch)
|
|
|
|
t.sendLatency.Describe(ch)
|
2015-09-16 07:22:18 -07:00
|
|
|
ch <- t.failedBatches.Desc()
|
|
|
|
ch <- t.failedSamples.Desc()
|
2014-07-23 10:55:33 -07:00
|
|
|
ch <- t.queueLength.Desc()
|
|
|
|
ch <- t.queueCapacity.Desc()
|
2014-06-18 10:43:15 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
// Collect implements prometheus.Collector.
|
2015-03-29 17:42:04 -07:00
|
|
|
func (t *StorageQueueManager) Collect(ch chan<- prometheus.Metric) {
|
2014-06-18 10:43:15 -07:00
|
|
|
t.samplesCount.Collect(ch)
|
|
|
|
t.sendLatency.Collect(ch)
|
2014-07-23 10:55:33 -07:00
|
|
|
t.queueLength.Set(float64(len(t.queue)))
|
2015-09-16 07:22:18 -07:00
|
|
|
ch <- t.failedBatches
|
|
|
|
ch <- t.failedSamples
|
2014-07-23 10:55:33 -07:00
|
|
|
ch <- t.queueLength
|
|
|
|
ch <- t.queueCapacity
|
2014-06-18 10:43:15 -07:00
|
|
|
}
|
|
|
|
|
2015-08-20 08:18:46 -07:00
|
|
|
func (t *StorageQueueManager) sendSamples(s model.Samples) {
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
t.sendSemaphore <- true
|
|
|
|
|
2016-05-19 06:30:09 -07:00
|
|
|
go func() {
|
|
|
|
defer func() {
|
|
|
|
<-t.sendSemaphore
|
|
|
|
}()
|
|
|
|
|
|
|
|
// Samples are sent to the remote storage on a best-effort basis. If a
|
|
|
|
// sample isn't sent correctly the first time, it's simply dropped on the
|
|
|
|
// floor.
|
|
|
|
begin := time.Now()
|
|
|
|
err := t.tsdb.Store(s)
|
|
|
|
duration := time.Since(begin) / time.Second
|
|
|
|
|
|
|
|
labelValue := success
|
|
|
|
if err != nil {
|
|
|
|
log.Warnf("error sending %d samples to remote storage: %s", len(s), err)
|
|
|
|
labelValue = failure
|
|
|
|
t.failedBatches.Inc()
|
|
|
|
t.failedSamples.Add(float64(len(s)))
|
|
|
|
}
|
|
|
|
t.samplesCount.WithLabelValues(labelValue).Add(float64(len(s)))
|
|
|
|
t.sendLatency.Observe(float64(duration))
|
|
|
|
}()
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// Run continuously sends samples to the remote storage.
|
|
|
|
func (t *StorageQueueManager) Run() {
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
defer func() {
|
|
|
|
close(t.drained)
|
|
|
|
}()
|
|
|
|
|
2015-03-29 17:42:04 -07:00
|
|
|
// Send batches of at most maxSamplesPerSend samples to the remote storage.
|
|
|
|
// If we have fewer samples than that, flush them out after a deadline
|
|
|
|
// anyways.
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case s, ok := <-t.queue:
|
|
|
|
if !ok {
|
2015-05-20 09:10:29 -07:00
|
|
|
log.Infof("Flushing %d samples to remote storage...", len(t.pendingSamples))
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
t.flush()
|
2015-05-20 09:10:29 -07:00
|
|
|
log.Infof("Done flushing.")
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
return
|
|
|
|
}
|
|
|
|
|
2015-03-14 19:36:15 -07:00
|
|
|
t.pendingSamples = append(t.pendingSamples, s)
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
|
|
|
|
for len(t.pendingSamples) >= maxSamplesPerSend {
|
2016-05-19 06:30:09 -07:00
|
|
|
t.sendSamples(t.pendingSamples[:maxSamplesPerSend])
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
t.pendingSamples = t.pendingSamples[maxSamplesPerSend:]
|
|
|
|
}
|
|
|
|
case <-time.After(batchSendDeadline):
|
|
|
|
t.flush()
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-02-14 10:36:27 -08:00
|
|
|
// Flush flushes remaining queued samples.
|
2015-03-29 17:42:04 -07:00
|
|
|
func (t *StorageQueueManager) flush() {
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
if len(t.pendingSamples) > 0 {
|
2016-05-19 06:30:09 -07:00
|
|
|
t.sendSamples(t.pendingSamples)
|
Add optional sample replication to OpenTSDB.
Prometheus needs long-term storage. Since we don't have enough resources
to build our own timeseries storage from scratch ontop of Riak,
Cassandra or a similar distributed datastore at the moment, we're
planning on using OpenTSDB as long-term storage for Prometheus. It's
data model is roughly compatible with that of Prometheus, with some
caveats.
As a first step, this adds write-only replication from Prometheus to
OpenTSDB, with the following things worth noting:
1)
I tried to keep the integration lightweight, meaning that anything
related to OpenTSDB is isolated to its own package and only main knows
about it (essentially it tees all samples to both the existing storage
and TSDB). It's not touching the existing TieredStorage at all to avoid
more complexity in that area. This might change in the future,
especially if we decide to implement a read path for OpenTSDB through
Prometheus as well.
2)
Backpressure while sending to OpenTSDB is handled by simply dropping
samples on the floor when the in-memory queue of samples destined for
OpenTSDB runs full. Prometheus also only attempts to send samples once,
rather than implementing a complex retry algorithm. Thus, replication to
OpenTSDB is best-effort for now. If needed, this may be extended in the
future.
3)
Samples are sent in batches of limited size to OpenTSDB. The optimal
batch size, timeout parameters, etc. may need to be adjusted in the
future.
4)
OpenTSDB has different rules for legal characters in tag (label) values.
While Prometheus allows any characters in label values, OpenTSDB limits
them to a to z, A to Z, 0 to 9, -, _, . and /. Currently any illegal
characters in Prometheus label values are simply replaced by an
underscore. Especially when integrating OpenTSDB with the read path in
Prometheus, we'll need to reconsider this: either we'll need to
introduce the same limitations for Prometheus labels or escape/encode
illegal characters in OpenTSDB in such a way that they are fully
decodable again when reading through Prometheus, so that corresponding
timeseries in both systems match in their labelsets.
Change-Id: I8394c9c55dbac3946a0fa497f566d5e6e2d600b5
2013-12-09 08:03:49 -08:00
|
|
|
}
|
|
|
|
t.pendingSamples = t.pendingSamples[:0]
|
|
|
|
}
|