2013-02-07 02:38:01 -08:00
|
|
|
// Copyright 2013 Prometheus Team
|
2012-11-26 11:11:34 -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.
|
|
|
|
|
2012-11-24 03:33:34 -08:00
|
|
|
package main
|
|
|
|
|
|
|
|
import (
|
2013-01-22 10:32:56 -08:00
|
|
|
"flag"
|
2013-06-25 05:02:27 -07:00
|
|
|
"os"
|
|
|
|
"os/signal"
|
2014-04-14 16:02:15 -07:00
|
|
|
"sync"
|
2013-08-14 03:36:55 -07:00
|
|
|
"syscall"
|
2013-06-25 05:02:27 -07:00
|
|
|
"time"
|
|
|
|
|
2013-08-12 08:18:02 -07:00
|
|
|
"github.com/golang/glog"
|
2013-06-25 05:02:27 -07:00
|
|
|
"github.com/prometheus/client_golang/extraction"
|
|
|
|
|
2013-08-13 07:14:18 -07:00
|
|
|
clientmodel "github.com/prometheus/client_golang/model"
|
2014-06-06 02:55:53 -07:00
|
|
|
registry "github.com/prometheus/client_golang/prometheus"
|
2013-08-13 07:14:18 -07:00
|
|
|
|
2013-01-27 09:49:45 -08:00
|
|
|
"github.com/prometheus/prometheus/config"
|
2013-07-30 08:18:07 -07:00
|
|
|
"github.com/prometheus/prometheus/notification"
|
2013-01-27 09:49:45 -08:00
|
|
|
"github.com/prometheus/prometheus/retrieval"
|
2014-05-28 10:44:54 -07:00
|
|
|
"github.com/prometheus/prometheus/rules/manager"
|
2014-06-06 02:55:53 -07:00
|
|
|
"github.com/prometheus/prometheus/storage/local"
|
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
|
|
|
"github.com/prometheus/prometheus/storage/remote"
|
|
|
|
"github.com/prometheus/prometheus/storage/remote/opentsdb"
|
2013-02-08 05:49:55 -08:00
|
|
|
"github.com/prometheus/prometheus/web"
|
2013-05-05 10:32:04 -07:00
|
|
|
"github.com/prometheus/prometheus/web/api"
|
2012-11-24 03:33:34 -08:00
|
|
|
)
|
|
|
|
|
2013-06-25 05:02:27 -07:00
|
|
|
const deletionBatchSize = 100
|
2013-05-13 01:53:24 -07:00
|
|
|
|
2013-01-22 10:32:56 -08:00
|
|
|
// Commandline flags.
|
|
|
|
var (
|
2013-08-13 04:58:19 -07:00
|
|
|
configFile = flag.String("configFile", "prometheus.conf", "Prometheus configuration file name.")
|
|
|
|
metricsStoragePath = flag.String("metricsStoragePath", "/tmp/metrics", "Base path for metrics storage.")
|
|
|
|
|
|
|
|
alertmanagerUrl = flag.String("alertmanager.url", "", "The URL of the alert manager to send notifications to.")
|
2013-08-13 04:43:37 -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
|
|
|
remoteTSDBUrl = flag.String("storage.remote.url", "", "The URL of the OpenTSDB instance to send samples to.")
|
|
|
|
remoteTSDBTimeout = flag.Duration("storage.remote.timeout", 30*time.Second, "The timeout to use when sending samples to OpenTSDB.")
|
|
|
|
|
2013-08-13 04:58:19 -07:00
|
|
|
samplesQueueCapacity = flag.Int("storage.queue.samplesCapacity", 4096, "The size of the unwritten samples queue.")
|
|
|
|
diskAppendQueueCapacity = flag.Int("storage.queue.diskAppendCapacity", 1000000, "The size of the queue for items that are pending writing to disk.")
|
|
|
|
memoryAppendQueueCapacity = flag.Int("storage.queue.memoryAppendCapacity", 10000, "The size of the queue for items that are pending writing to memory.")
|
2013-05-07 08:14:04 -07:00
|
|
|
|
2013-08-23 06:59:51 -07:00
|
|
|
deleteInterval = flag.Duration("delete.interval", 11*time.Hour, "The amount of time between deletion of old values.")
|
2013-05-13 01:53:24 -07:00
|
|
|
|
2013-08-23 06:59:51 -07:00
|
|
|
deleteAge = flag.Duration("delete.ageMaximum", 15*24*time.Hour, "The relative maximum age for values before they are deleted.")
|
2013-05-14 08:50:52 -07:00
|
|
|
|
2014-06-06 02:55:53 -07:00
|
|
|
memoryEvictionInterval = flag.Duration("storage.memory.evictionInterval", 15*time.Minute, "The period at which old data is evicted from memory.")
|
|
|
|
memoryRetentionPeriod = flag.Duration("storage.memory.retentionPeriod", time.Hour, "The period of time to retain in memory during evictions.")
|
|
|
|
|
|
|
|
storagePurgeInterval = flag.Duration("storage.purgeInterval", time.Hour, "How frequently to purge old data from the storage.")
|
|
|
|
storageRetentionPeriod = flag.Duration("storage.retentionPeriod", 15*24*time.Hour, "The period of time to retain in storage.")
|
2013-07-30 08:18:07 -07:00
|
|
|
|
|
|
|
notificationQueueCapacity = flag.Int("alertmanager.notificationQueueCapacity", 100, "The size of the queue for pending alert manager notifications.")
|
2013-08-13 04:52:24 -07:00
|
|
|
|
|
|
|
printVersion = flag.Bool("version", false, "print version information")
|
2013-01-22 10:32:56 -08:00
|
|
|
)
|
|
|
|
|
2013-04-29 02:17:56 -07:00
|
|
|
type prometheus struct {
|
2013-06-25 05:02:27 -07:00
|
|
|
unwrittenSamples chan *extraction.Result
|
2013-05-07 08:14:04 -07:00
|
|
|
|
2014-05-28 10:44:54 -07:00
|
|
|
ruleManager manager.RuleManager
|
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
|
|
|
targetManager retrieval.TargetManager
|
|
|
|
notifications chan notification.NotificationReqs
|
2014-06-06 02:55:53 -07:00
|
|
|
storage storage_ng.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
|
|
|
remoteTSDBQueue *remote.TSDBQueueManager
|
2013-08-13 08:19:13 -07:00
|
|
|
|
2014-04-14 16:02:15 -07:00
|
|
|
closeOnce sync.Once
|
2013-04-29 02:17:56 -07:00
|
|
|
}
|
|
|
|
|
2013-05-07 08:14:04 -07:00
|
|
|
func (p *prometheus) interruptHandler() {
|
2013-04-29 02:17:56 -07:00
|
|
|
notifier := make(chan os.Signal)
|
2013-08-14 03:36:55 -07:00
|
|
|
signal.Notify(notifier, os.Interrupt, syscall.SIGTERM)
|
2013-04-29 02:17:56 -07:00
|
|
|
|
|
|
|
<-notifier
|
|
|
|
|
2013-08-14 03:36:55 -07:00
|
|
|
glog.Warning("Received SIGINT/SIGTERM; Exiting gracefully...")
|
2014-04-14 16:02:15 -07:00
|
|
|
|
|
|
|
p.Close()
|
|
|
|
|
2013-04-29 02:17:56 -07:00
|
|
|
os.Exit(0)
|
|
|
|
}
|
|
|
|
|
2014-04-14 16:02:15 -07:00
|
|
|
func (p *prometheus) Close() {
|
|
|
|
p.closeOnce.Do(p.close)
|
|
|
|
}
|
|
|
|
|
2013-05-07 08:14:04 -07:00
|
|
|
func (p *prometheus) close() {
|
2014-04-14 16:02:15 -07:00
|
|
|
// The "Done" remarks are a misnomer for some subsystems due to lack of
|
|
|
|
// blocking and synchronization.
|
|
|
|
glog.Info("Shutdown has been requested; subsytems are closing:")
|
|
|
|
p.targetManager.Stop()
|
|
|
|
glog.Info("Remote Target Manager: Done")
|
|
|
|
p.ruleManager.Stop()
|
|
|
|
glog.Info("Rule Executor: Done")
|
|
|
|
|
2013-12-11 06:30:27 -08:00
|
|
|
close(p.unwrittenSamples)
|
|
|
|
|
2014-06-06 02:55:53 -07:00
|
|
|
if err := p.storage.Close(); err != nil {
|
|
|
|
glog.Error("Error closing local storage: ", err)
|
|
|
|
}
|
2014-04-14 16:02:15 -07:00
|
|
|
glog.Info("Local Storage: Done")
|
2013-07-30 08:18:07 -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
|
|
|
if p.remoteTSDBQueue != nil {
|
|
|
|
p.remoteTSDBQueue.Close()
|
2014-04-14 16:02:15 -07:00
|
|
|
glog.Info("Remote Storage: Done")
|
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
|
|
|
}
|
|
|
|
|
2013-07-30 08:18:07 -07:00
|
|
|
close(p.notifications)
|
2014-04-14 16:02:15 -07:00
|
|
|
glog.Info("Sundry Queues: Done")
|
|
|
|
glog.Info("See you next time!")
|
2013-04-29 02:17:56 -07:00
|
|
|
}
|
|
|
|
|
2012-11-24 03:33:34 -08:00
|
|
|
func main() {
|
2013-04-29 02:17:56 -07:00
|
|
|
// TODO(all): Future additions to main should be, where applicable, glumped
|
|
|
|
// into the prometheus struct above---at least where the scoping of the entire
|
|
|
|
// server is concerned.
|
2013-01-22 10:32:56 -08:00
|
|
|
flag.Parse()
|
2013-04-25 02:47:48 -07:00
|
|
|
|
2013-04-25 04:14:50 -07:00
|
|
|
versionInfoTmpl.Execute(os.Stdout, BuildInfo)
|
|
|
|
|
2013-04-25 02:47:48 -07:00
|
|
|
if *printVersion {
|
|
|
|
os.Exit(0)
|
|
|
|
}
|
|
|
|
|
2013-01-22 10:32:56 -08:00
|
|
|
conf, err := config.LoadFromFile(*configFile)
|
2013-01-07 14:24:26 -08:00
|
|
|
if err != nil {
|
2013-08-12 08:18:02 -07:00
|
|
|
glog.Fatalf("Error loading configuration from %s: %v", *configFile, err)
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
|
2014-06-06 02:55:53 -07:00
|
|
|
persistence, err := storage_ng.NewDiskPersistence(*metricsStoragePath, 1024)
|
2013-03-27 03:25:05 -07:00
|
|
|
if err != nil {
|
2014-06-06 02:55:53 -07:00
|
|
|
glog.Fatal("Error opening disk persistence: ", err)
|
2013-05-02 09:27:12 -07:00
|
|
|
}
|
2014-06-06 02:55:53 -07:00
|
|
|
|
|
|
|
o := &storage_ng.MemorySeriesStorageOptions{
|
|
|
|
Persistence: persistence,
|
|
|
|
MemoryEvictionInterval: *memoryEvictionInterval,
|
|
|
|
MemoryRetentionPeriod: *memoryRetentionPeriod,
|
|
|
|
PersistencePurgeInterval: *storagePurgeInterval,
|
|
|
|
PersistenceRetentionPeriod: *storageRetentionPeriod,
|
|
|
|
}
|
|
|
|
memStorage, err := storage_ng.NewMemorySeriesStorage(o)
|
|
|
|
if err != nil {
|
|
|
|
glog.Fatal("Error opening memory series storage: ", err)
|
|
|
|
}
|
|
|
|
//registry.MustRegister(memStorage)
|
2013-04-29 02:17:56 -07:00
|
|
|
|
2014-06-18 10:43:15 -07:00
|
|
|
var remoteTSDBQueue *remote.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
|
|
|
if *remoteTSDBUrl == "" {
|
|
|
|
glog.Warningf("No TSDB URL provided; not sending any samples to long-term storage")
|
|
|
|
} else {
|
|
|
|
openTSDB := opentsdb.NewClient(*remoteTSDBUrl, *remoteTSDBTimeout)
|
|
|
|
remoteTSDBQueue = remote.NewTSDBQueueManager(openTSDB, 512)
|
2014-06-18 10:43:15 -07:00
|
|
|
registry.MustRegister(remoteTSDBQueue)
|
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
|
|
|
go remoteTSDBQueue.Run()
|
|
|
|
}
|
|
|
|
|
2013-06-25 05:02:27 -07:00
|
|
|
unwrittenSamples := make(chan *extraction.Result, *samplesQueueCapacity)
|
2013-08-13 07:14:18 -07:00
|
|
|
ingester := &retrieval.MergeLabelsIngester{
|
|
|
|
Labels: conf.GlobalLabels(),
|
|
|
|
CollisionPrefix: clientmodel.ExporterLabelPrefix,
|
|
|
|
|
|
|
|
Ingester: retrieval.ChannelIngester(unwrittenSamples),
|
|
|
|
}
|
2014-02-19 07:03:30 -08:00
|
|
|
|
2013-05-05 10:32:04 -07:00
|
|
|
// Queue depth will need to be exposed
|
2014-07-29 11:31:11 -07:00
|
|
|
targetManager := retrieval.NewTargetManager(ingester)
|
2013-05-05 10:32:04 -07:00
|
|
|
targetManager.AddTargetsFromConfig(conf)
|
|
|
|
|
2013-08-09 10:32:55 -07:00
|
|
|
notifications := make(chan notification.NotificationReqs, *notificationQueueCapacity)
|
2013-07-30 08:18:07 -07:00
|
|
|
|
2013-06-11 02:00:55 -07:00
|
|
|
// Queue depth will need to be exposed
|
2014-05-28 10:44:54 -07:00
|
|
|
ruleManager := manager.NewRuleManager(&manager.RuleManagerOptions{
|
2013-08-20 06:42:06 -07:00
|
|
|
Results: unwrittenSamples,
|
|
|
|
Notifications: notifications,
|
|
|
|
EvaluationInterval: conf.EvaluationInterval(),
|
2014-06-06 02:55:53 -07:00
|
|
|
Storage: memStorage,
|
2013-08-20 06:42:06 -07:00
|
|
|
PrometheusUrl: web.MustBuildServerUrl(),
|
|
|
|
})
|
2013-08-12 08:18:02 -07:00
|
|
|
if err := ruleManager.AddRulesFromConfig(conf); err != nil {
|
2013-08-12 09:22:48 -07:00
|
|
|
glog.Fatal("Error loading rule files: ", err)
|
2013-06-11 02:00:55 -07:00
|
|
|
}
|
|
|
|
go ruleManager.Run()
|
|
|
|
|
2013-08-20 06:42:06 -07:00
|
|
|
notificationHandler := notification.NewNotificationHandler(*alertmanagerUrl, notifications)
|
2014-06-18 10:43:15 -07:00
|
|
|
registry.MustRegister(notificationHandler)
|
2013-07-30 08:18:07 -07:00
|
|
|
go notificationHandler.Run()
|
|
|
|
|
2013-05-14 02:21:27 -07:00
|
|
|
flags := map[string]string{}
|
|
|
|
|
|
|
|
flag.VisitAll(func(f *flag.Flag) {
|
|
|
|
flags[f.Name] = f.Value.String()
|
|
|
|
})
|
|
|
|
|
2013-08-13 08:19:13 -07:00
|
|
|
prometheusStatus := &web.PrometheusStatusHandler{
|
|
|
|
BuildInfo: BuildInfo,
|
|
|
|
Config: conf.String(),
|
|
|
|
RuleManager: ruleManager,
|
|
|
|
TargetPools: targetManager.Pools(),
|
|
|
|
Flags: flags,
|
|
|
|
Birth: time.Now(),
|
2013-05-05 10:32:04 -07:00
|
|
|
}
|
|
|
|
|
2013-06-13 07:10:05 -07:00
|
|
|
alertsHandler := &web.AlertsHandler{
|
|
|
|
RuleManager: ruleManager,
|
|
|
|
}
|
|
|
|
|
2014-05-28 10:44:54 -07:00
|
|
|
consolesHandler := &web.ConsolesHandler{
|
2014-06-06 02:55:53 -07:00
|
|
|
Storage: memStorage,
|
2013-05-14 02:21:27 -07:00
|
|
|
}
|
2013-05-05 10:32:04 -07:00
|
|
|
|
|
|
|
metricsService := &api.MetricsService{
|
|
|
|
Config: &conf,
|
|
|
|
TargetManager: targetManager,
|
2014-06-06 02:55:53 -07:00
|
|
|
Storage: memStorage,
|
2013-05-05 10:32:04 -07:00
|
|
|
}
|
|
|
|
|
2013-06-25 05:02:27 -07:00
|
|
|
prometheus := &prometheus{
|
|
|
|
unwrittenSamples: unwrittenSamples,
|
2013-05-13 01:53:24 -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
|
|
|
ruleManager: ruleManager,
|
|
|
|
targetManager: targetManager,
|
|
|
|
notifications: notifications,
|
2014-06-06 02:55:53 -07:00
|
|
|
storage: memStorage,
|
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
|
|
|
remoteTSDBQueue: remoteTSDBQueue,
|
2013-04-29 02:17:56 -07:00
|
|
|
}
|
2014-04-14 16:02:15 -07:00
|
|
|
defer prometheus.Close()
|
|
|
|
|
|
|
|
webService := &web.WebService{
|
2014-06-06 02:55:53 -07:00
|
|
|
StatusHandler: prometheusStatus,
|
|
|
|
MetricsHandler: metricsService,
|
|
|
|
ConsolesHandler: consolesHandler,
|
|
|
|
AlertsHandler: alertsHandler,
|
2014-04-14 16:02:15 -07:00
|
|
|
|
|
|
|
QuitDelegate: prometheus.Close,
|
|
|
|
}
|
2013-04-29 02:17:56 -07:00
|
|
|
|
2013-06-06 01:42:21 -07:00
|
|
|
storageStarted := make(chan bool)
|
2014-06-06 02:55:53 -07:00
|
|
|
go memStorage.Serve(storageStarted)
|
2013-06-06 01:42:21 -07:00
|
|
|
<-storageStarted
|
|
|
|
|
2013-04-29 02:17:56 -07:00
|
|
|
go prometheus.interruptHandler()
|
2012-12-11 11:46:16 -08:00
|
|
|
|
2013-05-05 10:32:04 -07:00
|
|
|
go func() {
|
|
|
|
err := webService.ServeForever()
|
|
|
|
if err != nil {
|
2013-08-12 08:18:02 -07:00
|
|
|
glog.Fatal(err)
|
2013-05-05 10:32:04 -07:00
|
|
|
}
|
|
|
|
}()
|
2013-01-04 08:55:58 -08:00
|
|
|
|
2013-04-29 02:17:56 -07:00
|
|
|
// TODO(all): Migrate this into prometheus.serve().
|
2013-06-25 05:02:27 -07:00
|
|
|
for block := range unwrittenSamples {
|
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 block.Err == nil && len(block.Samples) > 0 {
|
2014-06-06 02:55:53 -07:00
|
|
|
memStorage.AppendSamples(block.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
|
|
|
if remoteTSDBQueue != nil {
|
|
|
|
remoteTSDBQueue.Queue(block.Samples)
|
|
|
|
}
|
2012-12-25 04:50:36 -08:00
|
|
|
}
|
2012-11-24 03:33:34 -08:00
|
|
|
}
|
|
|
|
}
|