[
https://issues.apache.org/jira/browse/IMPALA-9046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17200436#comment-17200436
]
ASF subversion and git services commented on IMPALA-9046:
---------------------------------------------------------
Commit 155ffd64ee3562c36d5000de3a5102e05b393cc2 in impala's branch
refs/heads/master from Sahil Takiar
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=155ffd6 ]
IMPALA-9046: Profile counter that indicates if a JVM pause occurred
Adds a new section to the host profiles that includes JVM GC related
metrics. These metrics are taken from JMX and the JvmPauseMonitor.
The host profiles will now include a section like below:
JVM:
- GcCount: 19
- GcNumInfoThresholdExceeded: 0
- GcNumWarnThresholdExceeded: 0
- GcTimeMillis: 17s476ms
- GcTotalExtraSleepTimeMillis: 380
GcNumInfoThresholdExceeded, GcNumWarnThresholdExceeded, and
GcTotalExtraSleepTimeMillis are all taken from JvmPauseMonitor.
GcCount and GcTimeMillis are taken from JMX (specifically,
GarbageCollectorMXBean).
The counters themselves are derived from the impalad host-level metrics.
Changed the 'lock_' in JvmMetricCache (in memory-metrics.h) from a mutex
to a shared_mutex. Most accessors of the JvmMetricCache member variables
are read-only. A write only occurs lazily at most every second. This
should help reduce lock contention on JvmMetricCache now that all
queries will start accessing info stored by the JvmMetricCache.
Testing:
* Ran core tests
* Added a test that runs Java UDF, which triggers JVM GC
Change-Id: Idbaae2f9142b8be94532a0a147668a3d96091b0b
Reviewed-on: http://gerrit.cloudera.org:8080/16414
Reviewed-by: Tim Armstrong <[email protected]>
Reviewed-by: Sahil Takiar <[email protected]>
Tested-by: Sahil Takiar <[email protected]>
> Profile counter that indicates if a process or JVM pause occurred
> -----------------------------------------------------------------
>
> Key: IMPALA-9046
> URL: https://issues.apache.org/jira/browse/IMPALA-9046
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Affects Versions: Impala 3.3.0
> Reporter: Tim Armstrong
> Assignee: Tamas Mate
> Priority: Major
>
> We currently log a message if a process or JVM pause is detected but there's
> no indication in the query profile if it got affected. I suggest that we
> should:
> * Add metrics that indicate the number and duration of detected pauses
> * Add counters to the backend profile for the deltas in those metrics
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
