Statsd client sometimes drops metrics when this queueSize of statsd client with max unprocessed messages is completely full. This causes some high cardinality metrics like per partition lag being droppped.
There are multiple parameters of statsdclient that can be initialized and can help increase the load/capacity of client to not to drop metrics more frequently.
Properties like queueSize, poolSize, processorWorkers and senderWorkers will now be configurable at runtime
Changes:
Add the following indexer level task metrics:
- `worker/task/running/count`
- `worker/task/assigned/count`
- `worker/task/completed/count`
These metrics will provide more visibility into the tasks distribution across indexers
(We often see a task skew issue across indexers and with this issue it would be easier
to catch the imbalance)
Changes:
- Make ServiceMetricEvent.Builder extend ServiceEventBuilder<ServiceMetricEvent>
and thus convert it to a plain builder rather than a builder of builder.
- Add methods setCreatedTime , setMetricAndValue to the builder
### Description
Added the following metrics, which are calculated from the `KillUnusedSegments` coordinatorDuty
`"killTask/availableSlot/count"`: calculates the number remaining task slots available for auto kill
`"killTask/maxSlot/count"`: calculates the maximum number of tasks available for auto kill
`"killTask/task/count"`: calculates the number of tasks submitted by auto kill.
#### Release note
NEW: metrics added for auto kill
`"killTask/availableSlot/count"`: calculates the number remaining task slots available for auto kill
`"killTask/maxSlot/count"`: calculates the maximum number of tasks available for auto kill
`"killTask/task/count"`: calculates the number of tasks submitted by auto kill.
* merge druid-core, extendedset, and druid-hll into druid-processing to simplify everything
* fix poms and license stuff
* mockito is evil
* allow reset of JvmUtils RuntimeInfo if tests used static injection to override
Changes:
- Add a metric for partition-wise kafka/kinesis lag for streaming ingestion.
- Emit lag metrics for streaming ingestion when supervisor is not suspended and state is in {RUNNING, IDLE, UNHEALTHY_TASKS, UNHEALTHY_SUPERVISOR}
- Document metrics
It was found that the namespace/cache/heapSizeInBytes metric that tracks the total heap size in bytes of all lookup caches loaded on a service instance was being under reported. We were not accounting for the memory overhead of the String object, which I've found in testing to be ~40 bytes. While this overhead may be java version dependent, it should not vary much, and accounting for this provides a better estimate. Also fixed some logging, and reading bytes from the JDBI result set a little more efficient by saving hash table lookups. Also added some of the lookup metrics to the default statsD emitter metric whitelist.
* remove use of mocks for ServiceMetricEvent
* simplify KafkaEmitterTests by moving to Mockito
* speed up KafkaEmitterTest by adjusting reporting frequency in tests
* remove unnecessary easymock and JUnitParams dependencies
* Add worker category as dimension in TaskSlotCountStatsMonitor
* Change description
* Add workerConfig as field
* Modify HttpRemoteTaskRunnerTest to test worker category in taskslot metrics
* Fixing tests
* Fixing alerts
* Adding unit test in SingleTaskBackgroundRunnerTest for task slot metrics APIs
* Resolving false positive spell check
* addressing comments
* throw UnsupportedOperationException for tasklotmetrics APIs in SingleTaskBackgroundRunner
Co-authored-by: Nikhil Navadiya <nnavadiya@twitter.com>
* prometheus-emitter
* use existing jetty server to expose prometheus collection endpoint
* unused variables
* better variable names
* removed unused dependencies
* more metric definitions
* reorganize
* use prometheus HTTPServer instead of hooking into Jetty server
* temporary empty help string
* temporary non-empty help. fix incorrect dimension value in JSON (also updated statsd json)
* added full help text. added metric conversion factor for timers that are not using seconds. Correct metric dimension name in documentation
* added documentation for prometheus emitter
* safety for invalid labelNames
* fix travis checks
* Unit test and better sanitization of metrics names and label values
* add precondition to check namespace against regex
* use precompiled regex
* remove static imports. fix metric types
* better docs. fix possible NPE in PrometheusEmitterConfig. Guard against multiple calls to PrometheusEmitter.start()
* Update regex for label-value replacements to allow internal numeric values. Additional tests
* Adds missing license header
updates website/.spelling to add words used in prometheus-emitter docs.
updates docs/operations/metrics.md to correct the spelling of
bufferPoolName
* fixes version in extensions-contrib/prometheus-emitter
* fix style guide errors
* update import ordering
* add another word to website/.spelling
* remove unthrown declared exception
* remove unused import
* Pushgateway strategy for metrics
* typo
* Format fix and nullable strategy
* Update pom file for prometheus-emitter
* code review comments. Counter to gauge for cache metrics, periodical task to pushGateway
* Syntax fix
* Dimension label regex include numeric character back, fix previous commit
* bump prometheus-emitter pom dev version
* Remove scheduled task inside poen that push metrics
* Fix checkstyle
* Unit test coverage
* Unit test coverage
* Spelling
* Doc fix
* spelling
Co-authored-by: Michael Schiff <michael.schiff@tubemogul.com>
Co-authored-by: Michael Schiff <schiff.michael@gmail.com>
Co-authored-by: Tianxin Zhao <tianxin.zhao@tubemogul.com>
Co-authored-by: Tianxin Zhao <tizhao@adobe.com>
except success and failure stats, task count metrics should all be
gauges, since they represent the current state and not some aggregate
counter over time.
* Adding more worker metrics to Druid Overlord
* Changing the nomenclature from worker to peon as that represents the metrics that we want to monitor better
* Few more instance of worker usage replaced with peon
* Modifying the peon idle count logic to only use eligible workers available capacity
* Changing the naming to task slot count instead of peon
* Adding some unit test coverage for the new test runner apis
* Addressing Review Comments
* Modifying the TaskSlotCountStatsProvider apis so that overlords which are not leader do not emit these metrics
* Fixing the spelling issue in the docs
* Setting the annotation Nullable on the TaskSlotCountStatsProvider methods
* IntelliJ inspection and checkstyle rule for "Collection.EMPTY_* field accesses replaceable with Collections.empty*()"
* Reverted checkstyle rule
* Added tests to pass CI
* Codestyle
Adarsh Sanjeev
Hardik Bajaj
Rushikesh Bankar
YongGang
Karan Kumar
Laksh Singla
Tejaswini Bandlamudi
Abhishek Radhakrishnan
Kashif Faraz
zachjsh
Suneet Saldanha
AmatyaAvadhanula
Clint Wylie
Abhishek Agarwal
Didip Kerabat
dependabot[bot]
Xavier Léauté
Nikhil Navadiya
Jian Wang
Tianxin Zhao
Jihoon Son
Atul Mohan
Jonathan Wei
Mainak Ghosh
Aleksey Plekhanov
Gian Merlino
jon-wei