These are chat archives for ManageIQ/manageiq/performance

30th
Apr 2018
Pradeep Kumar Surisetty
@psuriset
Apr 30 2018 13:35
@kbrock @dmetzger57 ping
Keenan Brock
@kbrock
Apr 30 2018 13:36
how can we help?
Pradeep Kumar Surisetty
@psuriset
Apr 30 2018 13:37
i see lot of unused pgsql indexes on DB appliance http://pastebin.test.redhat.com/583871
is this a known stuff
This is 6K VMware setup + 3 worker + 1 DB appliance
@kbrock ^
Keenan Brock
@kbrock
Apr 30 2018 13:41
thnx - will look over list
dont think we have 100% knowledge of used indexes
Keenan Brock
@kbrock
Apr 30 2018 14:42

@psuriset that list of unused indexes looks pretty big / depressing
after you enabled the pg_stat extension, what backend workloads did you run through the app?
Did you run a script to exercise the ui as well?

just trying to figure out which ones are false positives.

Thanks for running this

Keenan Brock
@kbrock
Apr 30 2018 14:47
336 unused indexes... ouch

@psuriset what does this say?

Some user procedures does not have custom cost and rows settings :

  • public.metrics_inheritance_before
  • public.metrics_inheritance_after
  • public.metric_rollups_inheritance_before
  • public.metric_rollups_inheritance_after

@dmetzger57
A:

set vm.overcommit_memory=2 in /etc/sysctl.conf and run sysctl -p to reload it.
This will disable memory overcommitment and avoid postgresql killed by OOM killer.

B:

Configure your system to use noop or deadline io scheduler when on virtual machines :smile:

  • echo deadline > /sys/block/sdX/queue/scheduler
  • update your kernel parameters line with elevator=deadline to keep this parameter at next reboot
Keenan Brock
@kbrock
Apr 30 2018 14:53
@psuriset also. not sure what it means that pg_stat_statements is not enabled. Does that mean if the query is quicker than log_min_duration_statement that it will not mark that an index as used?
(thus the most efficient indexes won't get counted as used)