These are chat archives for ManageIQ/manageiq/performance

5th
Dec 2017
Dennis Metzger
@dmetzger57
Dec 05 2017 14:23
Started an appliance last night, idle (no provider), the Miq Server leak is present
IdleCfme.pdf
slow, but growing
Adam Grare
@agrare
Dec 05 2017 14:31
@dmetzger57 are these upstream or downstream appliances?
Dennis Metzger
@dmetzger57
Dec 05 2017 14:32
downstream (cfme)
Keenan Brock
@kbrock
Dec 05 2017 14:32
@dmetzger57 that didn't used to be the case (you tested ~1 month ago). what is the difference?
Dennis Metzger
@dmetzger57
Dec 05 2017 14:33
newer build ¯_(ツ)_/¯
Keenan Brock
@kbrock
Dec 05 2017 14:33
yea...
if we think it is a C gem thing, maybe we can compare gem lists
Adam Grare
@agrare
Dec 05 2017 14:35
@Fryguy and I are wondering if it might be SCL since we can only repo on an appliance
can you try with an upstream appliance?
Dennis Metzger
@dmetzger57
Dec 05 2017 14:36
started one this morning :smile:
Adam Grare
@agrare
Dec 05 2017 14:36
awesome
if its a C ext that rules out drb since that's pure ruby
maybe the pg gem?
Keenan Brock
@kbrock
Dec 05 2017 15:05
@agrare well, it is C - so that could be anything that leaks in c land
I wouldn't rule out drb ;)
Jason Frey
@Fryguy
Dec 05 2017 15:05
drb is written in Ruby, is @agrare's point
and we don't seem to be leaking threads (or more specifically any live objects)
Joe Rafaniello
@jrafanie
Dec 05 2017 15:13
@dmetzger57 can you DM me the IP, I'd like to look at the GC logging?
We need more examples showing the growing RSS/ OS heap size while ruby's live objects and heap pages are not growing
Dennis Metzger
@dmetzger57
Dec 05 2017 15:16
@jrafanie I’ll pm you the ip
Joe Rafaniello
@jrafanie
Dec 05 2017 15:17
:+1:
Joe Rafaniello
@jrafanie
Dec 05 2017 17:38
Nick LaMuro
@NickLaMuro
Dec 05 2017 17:40
I have a hard coded s/light reading/face melting read/ in my brain for you specifically
Joe Rafaniello
@jrafanie
Dec 05 2017 17:43
That means I did something right
Keenan Brock
@kbrock
Dec 05 2017 22:43
@NickLaMuro Was looking through queries and got frustrated - made a PR to shorten all our "vm".* where "type" in (......) queries
ManageIQ/manageiq#16602
Nick LaMuro
@NickLaMuro
Dec 05 2017 22:45
Looks neat, but unsure of the risk with it (this says more about my inexperience with the product, and not the change). That said, so far I like what it does.
Keenan Brock
@kbrock
Dec 05 2017 22:46
When we are profiling, it takes so much work to try and read them
the savings in just the time it takes for me to reformat the sql will be significant.
@NickLaMuro do you have any PRs you want to push? I picked off a few but...
Nick LaMuro
@NickLaMuro
Dec 05 2017 22:47
That said, the personal note section should either have something everyone knows how to use, or you should keep that somewhere else that isn't public </2cents>
Keenan Brock
@kbrock
Dec 05 2017 22:47
yea, these PRs take forever to merge. and come the time that I get back here, I can't find the files
maybe I can use an html comment
ugh. nope
Nick LaMuro
@NickLaMuro
Dec 05 2017 22:49
suggestion maybe: keep a folder of PR_NOTES in your ~/Documents folder and organize the sub folders for this kinda stuff by repo and PR number
Keenan Brock
@kbrock
Dec 05 2017 22:49
yea +1
Nick LaMuro
@NickLaMuro
Dec 05 2017 22:50
or... change it to something that doesn't need tooling that only use know how to setup
Keenan Brock
@kbrock
Dec 05 2017 22:55
@NickLaMuro how is valgrind?
Nick LaMuro
@NickLaMuro
Dec 05 2017 22:56
heh... a journey
will say, I do find cgdb pretty slick (versus vanilla gdb)
Keenan Brock
@kbrock
Dec 05 2017 22:56
any menial tasks you want to farm out?
aah
Nick LaMuro
@NickLaMuro
Dec 05 2017 22:58
no, I am basically getting a bug in ruby doing valgrind --leak-check=full -v ruby "puts 'foo'"...
and it is failing so far in startup, that the ruby bug report code is breaking because it is trying to reference a nil pointer in the VM object (ruby's VM, not our codebase)
so trying to work a patch to side step this, so I can get to the real error, and then maybe figure that out...
so there is that...
plus, this seems to just be an OSX issue (no surprise there I guess), but when trying to do this in the vagrant VM, it ran out of disk space on the default VM disk (other other VM this time...)
so it's pretty much picking between one rabbit hole and the other
Keenan Brock
@kbrock
Dec 05 2017 23:01
we're able to reproduce the leak though
that is good