These are chat archives for ManageIQ/manageiq/performance

26th
Aug 2015
Alex Krzos
@akrzos
Aug 26 2015 13:16
@dmetzger57 So ran the patched code with a VIM Broker as well, and I saw improvements on each provider size: Small - ~7%, Medium: ~26%, Large ~46%, XLarge 75%
On the X-Large the non-patched code took 3585s total on the best run (4 runs) and the best run with the patched code was 898s
Dennis Metzger
@dmetzger57
Aug 26 2015 13:38
@akrzos While not optimal, this is more acceptable.
Alex Krzos
@akrzos
Aug 26 2015 13:39
So thats only Initial Refreshes too so not sure if there is a subsequent improvement in targeted refreshes etc
vimbroker-patched.png
@dmetzger57 Thats VIMBroker Total timings
Also note I don't track the amount of time it took to "seed" the VIMBroker
Dennis Metzger
@dmetzger57
Aug 26 2015 13:41
the post initial refresh improvement is proportional to the number of VMs being linked, so the impact is tied to the number VMs that are new / added
Alex Krzos
@akrzos
Aug 26 2015 13:42
thats what i thought so I didn't run any targeted tests, but I can certainly run those as well
Keenan Brock
@kbrock
Aug 26 2015 13:44
@dmetzger57 I was looking at sam saffron debugging memory leaks and reminded of post by jnunemaker - would love to get this data into statsd/graphite/influxdb/whatever
Dennis Metzger
@dmetzger57
Aug 26 2015 13:45
@matthewd your bulk connect idea is a nice win for the XLarge environment
Keenan Brock
@kbrock
Aug 26 2015 13:45
seems odd that @akrzos would have to parse logs to get data that was already structured before we put into the logs in the first placec
I assume you've already read old post measure everything that kicked off the statsd kraze
Dennis Metzger
@dmetzger57
Aug 26 2015 13:47
no, but it's on my reading list now :-)
Alex Krzos
@akrzos
Aug 26 2015 13:47
So to track memory leaks, we would need a long term "stress" testbed
Keenan Brock
@kbrock
Aug 26 2015 13:48
etsy goes quite deep and kicked off a bunch of tools (in the hundereds)
well, not sure if we need to track memory leaks
Alex Krzos
@akrzos
Aug 26 2015 13:48
scale environment would be actually perfect for this
Keenan Brock
@kbrock
Aug 26 2015 13:48
just want to start tracking SOMETHING
Matthew Draper
@matthewd
Aug 26 2015 13:48
+1 statsd all the things
Keenan Brock
@kbrock
Aug 26 2015 13:48
want to see how many queries, how much data throughput, ... we do on the same test runs over time
disclaimer I say statsd but mean statsd / graphite / flavor of the month (which is influxdb for me right now)
but the idea is to make metric colletion so easy, that you start measuring everything. and then you can detect regressions, and when you performance tune, you can actually see things getting better
ala ruby bench