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"
|
|
|
|
|
|
|
|
"github.com/golang/glog"
|
|
|
|
|
|
|
|
clientmodel "github.com/prometheus/client_golang/model"
|
2014-06-18 10:43:15 -07:00
|
|
|
"github.com/prometheus/client_golang/prometheus"
|
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 (
|
|
|
|
// The maximum number of concurrent send requests to the TSDB.
|
|
|
|
maxConcurrentSends = 10
|
|
|
|
// The maximum number of samples to fit into a single request to the TSDB.
|
|
|
|
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"
|
|
|
|
)
|
|
|
|
|
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
|
|
|
// TSDBClient defines an interface for sending a batch of samples to an
|
|
|
|
// external timeseries database (TSDB).
|
|
|
|
type TSDBClient interface {
|
|
|
|
Store(clientmodel.Samples) error
|
|
|
|
}
|
|
|
|
|
|
|
|
// TSDBQueueManager manages a queue of samples to be sent to the TSDB indicated
|
|
|
|
// by the provided TSDBClient.
|
|
|
|
type TSDBQueueManager struct {
|
|
|
|
tsdb TSDBClient
|
|
|
|
queue chan clientmodel.Samples
|
|
|
|
pendingSamples clientmodel.Samples
|
|
|
|
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
|
|
|
|
sendErrors 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
|
|
|
}
|
|
|
|
|
2014-02-14 10:36:27 -08:00
|
|
|
// NewTSDBQueueManager builds a new TSDBQueueManager.
|
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
|
|
|
func NewTSDBQueueManager(tsdb TSDBClient, queueCapacity int) *TSDBQueueManager {
|
|
|
|
return &TSDBQueueManager{
|
|
|
|
tsdb: tsdb,
|
|
|
|
queue: make(chan clientmodel.Samples, queueCapacity),
|
|
|
|
sendSemaphore: make(chan bool, maxConcurrentSends),
|
|
|
|
drained: make(chan bool),
|
2014-06-18 10:43:15 -07:00
|
|
|
|
|
|
|
samplesCount: prometheus.NewCounterVec(
|
|
|
|
prometheus.CounterOpts{
|
2014-07-23 10:55:33 -07:00
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "sent_samples_total",
|
|
|
|
Help: "Total number of processed samples to be sent to remote TSDB.",
|
2014-06-18 10:43:15 -07:00
|
|
|
},
|
|
|
|
[]string{result},
|
|
|
|
),
|
2015-01-13 09:33:35 -08:00
|
|
|
sendLatency: prometheus.NewSummary(prometheus.SummaryOpts{
|
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "sent_latency_milliseconds",
|
|
|
|
Help: "Latency quantiles for sending sample batches to the remote TSDB.",
|
|
|
|
}),
|
|
|
|
sendErrors: prometheus.NewCounter(prometheus.CounterOpts{
|
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "sent_errors_total",
|
|
|
|
Help: "Total number of errors sending sample batches to the remote TSDB.",
|
|
|
|
}),
|
2014-07-23 10:55:33 -07:00
|
|
|
queueLength: prometheus.NewGauge(prometheus.GaugeOpts{
|
|
|
|
Namespace: namespace,
|
|
|
|
Subsystem: subsystem,
|
|
|
|
Name: "queue_length",
|
|
|
|
Help: "The number of processed samples queued to be sent to the remote TSDB.",
|
|
|
|
}),
|
|
|
|
queueCapacity: prometheus.MustNewConstMetric(
|
|
|
|
prometheus.NewDesc(
|
|
|
|
prometheus.BuildFQName(namespace, subsystem, "queue_capacity"),
|
|
|
|
"The capacity of the queue of samples to be sent to the remote TSDB.",
|
|
|
|
nil, nil,
|
|
|
|
),
|
|
|
|
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
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-02-14 10:36:27 -08:00
|
|
|
// Queue queues a sample batch to be sent to the TSDB. It drops the most
|
|
|
|
// recently queued samples on the floor if the queue is full.
|
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
|
|
|
func (t *TSDBQueueManager) Queue(s clientmodel.Samples) {
|
2015-02-12 10:47:24 -08:00
|
|
|
if len(s) == 0 {
|
|
|
|
return
|
|
|
|
}
|
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:
|
2014-06-18 10:43:15 -07:00
|
|
|
t.samplesCount.WithLabelValues(dropped).Add(float64(len(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
|
|
|
glog.Warningf("TSDB queue full, discarding %d samples", len(s))
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-10-10 05:19:02 -07:00
|
|
|
// Stop stops sending samples to the TSDB and waits for pending sends to
|
2014-06-18 10:43:15 -07:00
|
|
|
// complete.
|
2014-10-10 05:19:02 -07:00
|
|
|
func (t *TSDBQueueManager) Stop() {
|
2014-11-11 15:38:28 -08:00
|
|
|
glog.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
|
|
|
|
}
|
2014-11-11 15:38:28 -08:00
|
|
|
glog.Info("Remote storage stopped.")
|
2014-06-18 10:43:15 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
// Describe implements prometheus.Collector.
|
|
|
|
func (t *TSDBQueueManager) Describe(ch chan<- *prometheus.Desc) {
|
|
|
|
t.samplesCount.Describe(ch)
|
|
|
|
t.sendLatency.Describe(ch)
|
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.
|
|
|
|
func (t *TSDBQueueManager) Collect(ch chan<- prometheus.Metric) {
|
|
|
|
t.samplesCount.Collect(ch)
|
|
|
|
t.sendLatency.Collect(ch)
|
2014-07-23 10:55:33 -07:00
|
|
|
t.queueLength.Set(float64(len(t.queue)))
|
|
|
|
ch <- t.queueLength
|
|
|
|
ch <- t.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
|
|
|
func (t *TSDBQueueManager) sendSamples(s clientmodel.Samples) {
|
|
|
|
t.sendSemaphore <- true
|
|
|
|
defer func() {
|
|
|
|
<-t.sendSemaphore
|
|
|
|
}()
|
|
|
|
|
|
|
|
// Samples are sent to the TSDB 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)
|
2014-06-18 10:43:15 -07:00
|
|
|
duration := time.Since(begin) / time.Millisecond
|
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
|
|
|
|
2014-06-18 10:43:15 -07:00
|
|
|
labelValue := success
|
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 err != nil {
|
|
|
|
glog.Warningf("error sending %d samples to TSDB: %s", len(s), err)
|
2014-06-18 10:43:15 -07:00
|
|
|
labelValue = failure
|
2015-01-13 09:33:35 -08:00
|
|
|
t.sendErrors.Inc()
|
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
|
|
|
}
|
2014-06-18 10:43:15 -07:00
|
|
|
t.samplesCount.WithLabelValues(labelValue).Add(float64(len(s)))
|
2015-01-13 09:33:35 -08:00
|
|
|
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
|
|
|
}
|
|
|
|
|
2014-02-14 10:36:27 -08:00
|
|
|
// Run continuously sends samples to the TSDB.
|
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
|
|
|
func (t *TSDBQueueManager) Run() {
|
|
|
|
defer func() {
|
|
|
|
close(t.drained)
|
|
|
|
}()
|
|
|
|
|
|
|
|
// Send batches of at most maxSamplesPerSend samples to the TSDB. If we
|
|
|
|
// have fewer samples than that, flush them out after a deadline anyways.
|
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case s, ok := <-t.queue:
|
|
|
|
if !ok {
|
|
|
|
glog.Infof("Flushing %d samples to OpenTSDB...", len(t.pendingSamples))
|
|
|
|
t.flush()
|
|
|
|
glog.Infof("Done flushing.")
|
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
t.pendingSamples = append(t.pendingSamples, s...)
|
|
|
|
|
|
|
|
for len(t.pendingSamples) >= maxSamplesPerSend {
|
|
|
|
go t.sendSamples(t.pendingSamples[:maxSamplesPerSend])
|
|
|
|
t.pendingSamples = t.pendingSamples[maxSamplesPerSend:]
|
|
|
|
}
|
|
|
|
case <-time.After(batchSendDeadline):
|
|
|
|
t.flush()
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-02-14 10:36:27 -08:00
|
|
|
// Flush flushes remaining queued 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
|
|
|
func (t *TSDBQueueManager) flush() {
|
|
|
|
if len(t.pendingSamples) > 0 {
|
|
|
|
go t.sendSamples(t.pendingSamples)
|
|
|
|
}
|
|
|
|
t.pendingSamples = t.pendingSamples[:0]
|
|
|
|
}
|