mozilla :: #perf

7 Sep 2017
09:07jmaher|afkigoldan: for the enable tests by default on stylo, I just commented in the bug since we had switched the config over
09:33igoldanjmaher|afk: michal, on bug 1392841, is asking if the tp6-e10s suite can be run locally
09:34igoldanjmaher|afk: are there any difficulties with this?
09:38jmaher|afkigoldan: it should run locally with |mach talos-test tp6-e10s|
09:46igoldanjmaher|afk: I need to do retriggers & backfills for some big AWSY regressions
09:46igoldanjmaher|afk: but I noticed the big load on talos-linux64-ix machine
09:47igoldanjmaher|afk: can I continue with the retriggers?
09:49jmaher|afkigoldan: I would do the retriggers
09:49jmaher|afkwe need to wait for the machines to start working
09:50jmaher|afkbut that puts it in the queue
09:50* jmaher|afk goes afk for a bit
13:24jmaherigoldan: oh, the pgo build is finished, Idid some retriggers for you; results in <1 hour :)
13:25igoldanjmaher: ok
13:39jmaherigoldan: question- does your patch actually turn on mainthreadIO, if you removed some lines from the whitelist would you get a failure?
13:40jmaherigoldan: your patch doesn&#39;t seem to &#39;turn it on&#39;, just add more entries
13:42igoldanjmaher: for most of the files mentioned in mtio_whitelist, if you remove them, you should get failure
13:42igoldanjmaher: but I added some extra ones
13:43igoldanjmaher: before this change, I ran the old code, and grabbed the list with all the accessed files
13:43igoldanjmaher: I then made a push with only the files with fixed names, and no xperf errors came out
13:44igoldanjmaher: but I still wanted to add the extra files, with variable names, for extra future proof
13:45igoldanif you look at this push: https://treeherder.mozilla.org/#/jobs?repo=try&revision=01c9c537a6bb039b595ed5000c11879a7ce99383
13:46igoldanhere I only included in mtio_whitelist the files with fixed names
13:46igoldanand multiple xperf tests passed
13:47jmaherigoldan: but if it is not enabled now by default, where in the code are you enabling it?
13:47igoldanthis is the Try push from which I grabbed the list of file: https://treeherder.mozilla.org/#/jobs?repo=try&revision=03e9a9ba1abcd16ea779d6a88b14b20fa669afc7&selectedJob=127406055
13:47jmaherok, those failures look right
13:48jmaherthe change you have there is https://hg.mozilla.org/try/rev/03e9a9ba1abcd16ea779d6a88b14b20fa669afc7
13:48jmaherI don&#39;t see that in the patch I am reviewing
13:50igoldanjmaher: you are right...
13:51igoldanjmaher: I slipped this somehow
13:51jmaherthat is ok- glad we caught it now!
13:51jmaherI overlooked it yesterday and so far this morning
13:52igoldanthat&#39;s why the tests passed with only a subset of the files
13:53igoldanI&#39;m pushing to Try, to see if all goes well
13:54jmaherglad we are getting closer :)
13:57igoldanthanks for looking over this
14:00igoldancan you help me with this Try push?
14:00igoldanhttps://treeherder.mozilla.org/#/jobs?repo=try&revision=dae873214f718b5b0f3b93287cf7bb0bcc42c4e0&selectedJob=129240686
14:00igoldanI pushed using |mach try fuzzy|
14:01jmaherigoldan: ok, not sure what ba is
14:01jmahermaybe build artifact
14:01igoldanI think I should cancel these jobs?
14:01igoldanright?
14:01jmaherlets ask ahal
14:01igoldanI see try fuzzy has --no-artifact flag
14:01igoldanok
14:02jmaherthere is artifact builds, I am sure that is what you have
14:10igoldanshould I cancel these?
14:11igoldanand use full builds, instead?
14:11igoldanjmaher:
14:14jmaherigoldan: full builds are not necessary
14:14igoldanjmaher: ok
14:14igoldanjmaher: anyway, like the try fuzzy tool
14:15jmaherlet :ahal know in the pi channel :)
14:36jmaherigoldan: ok, a few failures on your try push- but not a big list
14:36igoldanjmaher: I see there&#39;s was an unexpected file accessed
14:37igoldanjmaher: I&#39;m adding that too to the list, and pushing again
14:41jmaheryeah, there were a few of them, look on pgo- a bit more than opt
15:29igoldanjmaher: yep
15:30igoldanjmaher: had to do some extra code changes: https://treeherder.mozilla.org/#/jobs?repo=try&revision=538678301a4385afad800b0783b02081d8af629d&selectedJob=129263213
15:31jmaherigoldan: ah, the ordereddict
15:31igoldanjmaher: yeah
15:31igoldanjmaher: I really don&#39;t like those uuids
15:31igoldanjmaher: they can easily re-disable mainthreadio
15:32jmaherwell, that is a good solution to avoid them
15:37igoldanjmaher: I&#39;m leaving now
15:37igoldanjmaher: have a nice day
15:37igoldanjmaher: and see you tomorrow!
15:42jmaherbye!
16:08digitaraldjmaher: are there plans to report on regressions on subsets of metrics, like DAMP?
16:09jmaherdigitarald: no
16:10jmaherdigitarald: the volume would be very large
16:10digitaraldjmaher: I assume you know https://chromeperf.appspot.com/ (navigation broken in Firefox)
16:12jmaherdigitarald: I have seen that in the past, but honestly I haven&#39;t spent much time on it
16:14jmaherdigitarald: we could generate a lot of data, but it wouldn&#39;t get used; our flaw is in inaccurate blaming of patches, which is more an issue of not being able to run a test on every single commit
16:18digitaraldjmaher|afk: chrome has the same issue, right
16:53jmaherdigitarald: I believe they have better ownership and a larger pool of sheriffs
16:54jmaherdigitarald: likewise the abiliity to run their tests more consistently
20:00digitaraldjmaher: can you r+, Ben seems out: https://github.com/mozilla/arewefastyet/pull/173
20:06jmaherdigitarald: I am not sure how to do that
20:06jmaheris there a review button?
20:07jmaherI am not too familiar with github, but it looks like ben is requested for review and I don&#39;t see options to take it over
20:08digitaraldjmaher: if you say &quot;r+&quot; here I merge it :)
20:09jmaherok, I did it in a comment on the PR
20:11digitaraldjmaher: thank you!
20:12jmaherno problem
20:12jmaherthanks for updating the test
20:57digitaraldjmaher|afk: can you kick off another run on https://arewefastyet.com/#machine=36&view=breakdown&suite=speedometer-misc ?
8 Sep 2017
No messages
   
Last message: 12 days and 18 hours ago