On all SearchHead cluster members with ver 8.0.2, every day we are observing that CPU utilization grows. After roughly two days CPU load grapsh looks like "climbing".
After our analysis we found that several queries are "zombied" and it looks like Splunk does not control them.
These processes runs on Operating System level endlessly like consuming more and more CPU over time.
In UI there is message that "Search auto-canceled"
Always on the end of search.log for such process we see ;
09-28-2020 14:52:57.907 INFO ReducePhaseExecutor - ReducePhaseExecutor=1 action=CANCEL 09-28-2020 14:52:57.907 INFO DispatchExecutor - User applied action=CANCEL while status=3 09-28-2020 14:52:58.906 INFO ReducePhaseExecutor - ReducePhaseExecutor=1 action=CANCEL 09-28-2020 14:52:58.906 INFO DispatchExecutor - User applied action=CANCEL while status=3 09-28-2020 14:52:59.906 INFO ReducePhaseExecutor - ReducePhaseExecutor=1 action=CANCEL |
Please help.
There's a known issue for the version - there's a deadlock situation happening for the version and causing the symptoms of long running searches with the never ending messages you have found.
09-28-2020 14:52:53.906 INFO DispatchExecutor - User applied action=CANCEL while status=3
09-28-2020 14:52:54.906 INFO ReducePhaseExecutor - ReducePhaseExecutor=1 action=CANCEL
This has been fixed in 8.0.2.1 and 8.0.3 and also you can add this as a work-around which may have minor search performance impact until you upgrade to the fixed versions.
** work-around
limits.conf on all SH.
[search]
remote_timeline= 0
** fixed versions: 8.0.2.1 and 8.0.3+
There's a known issue for the version - there's a deadlock situation happening for the version and causing the symptoms of long running searches with the never ending messages you have found.
09-28-2020 14:52:53.906 INFO DispatchExecutor - User applied action=CANCEL while status=3
09-28-2020 14:52:54.906 INFO ReducePhaseExecutor - ReducePhaseExecutor=1 action=CANCEL
This has been fixed in 8.0.2.1 and 8.0.3 and also you can add this as a work-around which may have minor search performance impact until you upgrade to the fixed versions.
** work-around
limits.conf on all SH.
[search]
remote_timeline= 0
** fixed versions: 8.0.2.1 and 8.0.3+
It's happening to me on version 8.0.3 right now.