These are chat archives for ManageIQ/manageiq/performance

26th
May 2016
Joe Rafaniello
@jrafanie
May 26 2016 03:52
bundler/bundler#4618, if I can get the tests for that PR to pass and they merge, that should save ~10 MB per process on future builds with a release of bundler having that change
Mooli Tayer
@moolitayer
May 26 2016 14:11
Could someone review #8958 pleaes? I was asked to review it but I don't know anything about vim_performance_daily.rb
Sorry ManageIQ/manageiq#8958
Joe Rafaniello
@jrafanie
May 26 2016 14:21
I hope this image summarizes the findings from yesterday RE: memory changes per version because of bundler...
memory_by_version.png
I am working on getting the bundler PR 4618 merged, it's not a huge memory savings but 8-10 MB per process on 5.5 is not minor
Also, I need to figure out the RIGHT arrow, why it jumped in the first place in 1.11.2
Additionally, I will probably need to work with @Fryguy to look at what changed in the Gemfiles/code from 5.5.4 -> 5.6 to see if there's low hanging :grapes: :apple: :banana: to decrease overall memory from other non-bundler things
Joe Rafaniello
@jrafanie
May 26 2016 14:28
@moolitayer you probably want to ask that question re 8958 in core... I commented on the PR... I have no idea how that code works but the PR doesn't explain why the change fixes it.
At least, I can't understand why at a glance
Joe Rafaniello
@jrafanie
May 26 2016 14:35
@Fryguy Let me know if you want to pair on :point_up: May 26, 2016 10:21 AM
Alex Krzos
@akrzos
May 26 2016 14:37
Sorry guys
Joe Rafaniello
@jrafanie
May 26 2016 14:39
Cool, thanks @akrzos
to be clear, is that UI worker actually running the automate code synchronously?
because we fixed a thread leak in automate execution recently
Alex Krzos
@akrzos
May 26 2016 14:41
could be the same as the other thread leak
this is on 5.5.4.0
so perhaps the same patch for https://bugzilla.redhat.com/show_bug.cgi?id=1321934 would fix it?
Joe Rafaniello
@jrafanie
May 26 2016 14:41
let me see what the fix was to see if it was in that tag
ManageIQ/manageiq#8399
ManageIQ/manageiq#8419
Alex Krzos
@akrzos
May 26 2016 14:45
Screen Shot 2016-05-26 at 10.44.55.png
Joe Rafaniello
@jrafanie
May 26 2016 14:45
Awesome, 5.5.4.0 doesn't have that fix
Alex Krzos
@akrzos
May 26 2016 14:45
yeah neither does 5.5.4.2
Joe Rafaniello
@jrafanie
May 26 2016 14:46
yup
my guess is it's running the same code the generic/priority automate worker was running synchronously in the UI worker
Can you apply those changes to confirm?
how easy is it to recreate?
Alex Krzos
@akrzos
May 26 2016 14:47
Pretty easy to recreate this issue right now iirc that patch is like 1 or 2 lines so should be easy to apply it
Joe Rafaniello
@jrafanie
May 26 2016 14:48
the second one is nice to have but the first is enough to fix the thread leak
Alex Krzos
@akrzos
May 26 2016 14:50
ok let me try that to confirm if that fixes at least the thread count
I just looked on my 5.5.4.2 appliance that ran through a provisioning workload with ordering something off the service catalog it does not display this issue so I assume it has to do with the complexity of the service catalog item that is ordered thus the need for the database to reproduce that issue
but off to testing that patch
Joe Rafaniello
@jrafanie
May 26 2016 14:55
oh man, git is awesome
git show 5.5.4.2:lib/miq_automation_engine/engine/miq_ae_method.rb |grep -C "5" self.teardown_drb_for_ruby_method
Alex Krzos
@akrzos
May 26 2016 15:03
poof goes the thread leak
Joe Rafaniello
@jrafanie
May 26 2016 15:03
@akrzos :wink2:
Alex Krzos
@akrzos
May 26 2016 15:05
yup all three places where the extra threads were showing up is not growing the threads now
Screen Shot 2016-05-26 at 11.06.02.png
Dennis Metzger
@dmetzger57
May 26 2016 15:23
can we clone LJ?
Keenan Brock
@kbrock
May 26 2016 15:23
only 1 way to find out
Joe Rafaniello
@jrafanie
May 26 2016 16:02
:wink:
Joe Rafaniello
@jrafanie
May 26 2016 18:35
@akrzos smem should be on 5.6 builds
I'll DM you