These are chat archives for ManageIQ/manageiq/performance

10th
Nov 2015
Alex Krzos
@akrzos
Nov 10 2015 01:24
@kbrock Bad news with that bz: process memory usage [5380087808] exceeded limit [5368709120], requesting worker to exit", :type=>"MiqPriorityWorker"}]
Worker worked it's way up to 5GB of memory
Keenan Brock
@kbrock
Nov 10 2015 01:24
that is not good :(
Joe Rafaniello
@jrafanie
Nov 10 2015 01:25
I see opportunities 😜
Alex Krzos
@akrzos
Nov 10 2015 01:25
if you tried to view grafana of that testbed over the weekend, it was a mess, my rhevm hosted-engine decided to quit, but it's all fixed now, so that scale setup is back up
another opportunity there is multiple workers exceeding the worker count
I believe, I have to double check to see if this is also reproducing that
Keenan Brock
@kbrock
Nov 10 2015 01:27
I'm having trouble running cap and U
is there a trick to setting "enable C&U" for a server?
yea
Alex Krzos
@akrzos
Nov 10 2015 01:28
My guess would be missing the region setting
Keenan Brock
@kbrock
Nov 10 2015 01:28
aah - cool
we need basic instructions how to run each of the server components
Alex Krzos
@akrzos
Nov 10 2015 01:28
under configure -configuration on the right side accordion, click the region
Keenan Brock
@kbrock
Nov 10 2015 01:28
so many people on our team don't know how to run basic stuff (me included)
thnx
Alex Krzos
@akrzos
Nov 10 2015 01:29
then look for C&U Collection tab
and check the boxes for collect on all clusters and all datastores to get everything
Keenan Brock
@kbrock
Nov 10 2015 01:46
ExtManagementSystem.first.perf_capture_enabled?
that shows false
had no idea that there was C&U settings under region
score -restart did it. thanks for all the help
Alex Krzos
@akrzos
Nov 10 2015 01:48
shows true for my scale setup
np
Jason Frey
@Fryguy
Nov 10 2015 01:50
I cheat and just do Host.all.each { |h| h.perf_capture_enabled = true }
Then repeat for EmsCluster and Storage
I think the UI is way too complicated there personally
Keenan Brock
@kbrock
Nov 10 2015 18:30
@matthewd any thoughts on counting queries using ActiveSupport::LogSubscriber#sql vs ActiveSupport::Notifications.subscribed(..., 'sql.active_record') ?
was comparing sql_queries_count gem vs our implementation of QueryCounter ala so
Matthew Draper
@matthewd
Nov 10 2015 18:45
Not really. Dunno whether they treat "schema" queries differently? Probably doesn't matter ¯\_(ツ)_/¯
Jason Frey
@Fryguy
Nov 10 2015 18:48
I already have that
see spec/support/query_counter.rb
or rather...it already uses the notifications
Keenan Brock
@kbrock
Nov 10 2015 18:48
thanks - found out they call the same type of thing
Jason Frey
@Fryguy
Nov 10 2015 18:48
gotcha
Keenan Brock
@kbrock
Nov 10 2015 18:48
subscribed vs subscribe == very cool
Jason Frey
@Fryguy
Nov 10 2015 18:49
yeah, that's what I used to generate https://gist.github.com/Fryguy/ec432fb7ed07767f8491
Still considerig making that a regular part of the CI runs
Keenan Brock
@kbrock
Nov 10 2015 18:49
I'm curious about adding GC counts and others like that
Jason Frey
@Fryguy
Nov 10 2015 18:50
interesting
Keenan Brock
@kbrock
Nov 10 2015 18:50
putting in "book ends" around the C&U collection
Jason Frey
@Fryguy
Nov 10 2015 18:50
yeah
Keenan Brock
@kbrock
Nov 10 2015 18:50
a little curious about whether we care about cache too
if there are too many cache hits...
@Fryguy what is the reasoning behind the true in MiqServer.my_server.zone(true)
we sure look up that zone lots of times
for 100VMs we look it up hundreds - if not thousands of times
that one object won't change the world, but curious the reasoning, to understand the other more substantial queries / caching strategy
Joe Rafaniello
@jrafanie
Nov 10 2015 18:55
@kbrock do you have a measurement or is that a guess?
Keenan Brock
@kbrock
Nov 10 2015 18:55
I was watching the logs go by - each block seemed to look up that object 2-3 times per host
I'm not changing anything
just starting to get numbers
but curious why we would need to reload it
not asking for that exact line, just for why in general
so I can take that in consideration while collecting / making hypothesis
Matthew Draper
@matthewd
Nov 10 2015 18:57
@kbrock well, you hardly want false zones, do you? :trollface:
Keenan Brock
@kbrock
Nov 10 2015 18:57
didn't realize "my zone" would change
and 1/2 way through a run, even if it did change, seems you'd want to just finish up before changing that
Jason Frey
@Fryguy
Nov 10 2015 20:34
I can't really say...depends where in the code it is
offhand seems silly, but it might have a reason, or the code in question is written generically and can't know