mozilla :: #perf

21 Apr 2017
08:45igoldanjmaher|afk: good morning!
08:48jmaher|afkigoldan: hi
08:48igoldanjmaher|afk: I want to know if this is the way you push backouts to Try
08:48igoldanjmaher|afk: https://wiki.mozilla.org/Sheriffing/How:To:Backouts
08:50jmaher|afkyes, but not thefull set
08:50jmaher|afkin this case you would push to try with your tree
08:50jmaher|afkthen do the hg qbackout to create the backout in a mq series- then push to try again
08:51igoldanjmaher|afk: after you run hg qbackout
08:52igoldanjmaher|afk: do you run hg qfinish -a also?
08:52igoldanjmaher|afk: and then push to Try?
08:53jmaher|afkno
08:53igoldanjmaher|afk: ok
08:54jmaher|afkI typically do: |hg qnew backout_<rev>; hg backout <rev>; hg qrefresh; ./mach try -b o -p <platform> -u none -t ... --rebuild 5|
08:55igoldanjmaher|afk: so you&#39;re not using qbackout, but the builtin backout
08:55jmaher|afkI do, but it is lack of familiarity- since I don&#39;t want to accidentally land anything :)
08:55igoldanjmaher|afk: I am aware there are some approaches/variations to doing backouts
08:58igoldanjmaher|afk: I&#39;ll later ask to see if I did the right thing for https://bugzilla.mozilla.org/show_bug.cgi?id=1358042
08:58jmaher|afkigoldan: ok- as a note, I will be on holiday today starting in <30 minutes :)
09:02igoldanjmaher|afk: how cool! you&#39;ll be out after weekend also?
09:08jmaher|afkigoldan: just today/tomorrow not around
09:08jmaher|afkso you will see me Monday as normal :)
09:09igoldanjmaher|afk: then I wish you a nice mini-holiday
09:11jmaher|afkigoldan: thanks!
09:11jmaher|afkigoldan: you have a few software fixes underway, are you doing ok with them? any concerns?
09:15jmaher|afkigoldan: I want to make sure you are not frustrated or overloaded
09:20igoldanjmaher|afk: I&#39;m looking over the https://bugzilla.mozilla.org/show_bug.cgi?id=1355934
09:20igoldanjmaher|afk: I reviewed the current definition of the tests
09:20igoldanjmaher|afk: that derive from talos.Test
09:21igoldanjmaher|afk: noticed some of the diff between TsTest and PageloaderTest
09:21jmaher|afkcool, that is learning a lot :)
09:21igoldanjmaher|afk: but I need to review the higher calling code
09:22jmaher|afkok, for the most part it is just moving files and updating a few things
09:23igoldanjmaher|afk: could you tell me, in advance, what will the manifest file look like?
09:23igoldanjmaher|afk: will it point to the html page that tpaint currently uses?
09:26jmaher|afkyes, that is it
09:27jmaher|afkand I will let you figure out if it needs a % or not :)
09:28igoldanjmaher|afk: another question: if I push to Try using a baseline using &quot;hg push try&quot; (for https://bugzilla.mozilla.org/show_bug.cgi?id=1358042)
09:28igoldanjmaher|afk: without the &quot;-f&quot; flag, I get a warning &quot;abort: source has mq patches applied&quot;
09:29igoldanjmaher|afk: is this normal?
09:29igoldanjmaher|afk: do you always use the &quot;-f&quot; flag when pushing to Try?
09:30jmaher|afkigoldan: yes- that is normal with a mq applied you need -f
09:30jmaher|afkbut that is try only- luckily that is all you need to push to
09:30igoldanjmaher|afk: great! thanks!
09:31igoldanjmaher|afk: yep
09:31igoldanjmaher|afk: these are all my questions for today
09:32jmaher|afkigoldan: awesome, thanks for working on this!
14:45gfritzscheehsan: hey, re bug 1358259.... i&#39;m concerned that this might imply that histogram recording is broken right now... should we just back out bug 1350765 until we know more?
14:45gfritzschehappy to help digging on monday FWIW
18:12mstangedthayer: ping
18:12dthayermstange: pong
18:13mstangehi
18:13dthayerhey!
18:13mstangeI&#39;m confused by the profile you linked to
18:13dthayerdid you get the second one?
18:13mstangeoh, I did not, one sec
18:13mstangeah, much better!
18:13mstangethat addresses my confusion :)
18:14dthayerhaha, yeah - I had put something in an inner loop that shouldn&#39;t have been there :P
18:17mstangedthayer: this profile is amazing
18:17dthayerhooray! glad to hear it.
18:18mstangedthayer: how many samples were there in the profile that hung my system?
18:20mstangedthayer: how does it happen that the same stack shows up multiple times in different places in the time line?
18:20dthayerseems 2002665
18:20mstangeooohhh okay
18:20mstangethat&#39;s about 100x the load that we usually deal with
18:20dthayerah, sorry I forgot to note, they&#39;re split out by day, and the last profile is actually the last 30 days
18:21mstangeah, I see
18:21mstangeinteresting
18:21mstangehave we had native stacks for this long already?
18:21mstangeI had no idea
18:22mstangethere&#39;s a lot of variance between the days
18:22dthayeryeah, I think for a couple months?
18:22dthayeryeah, i noticed that when I was just doing pseudo stacks too... couldn&#39;t find any patterns, but I didn&#39;t look too hard
18:22mstangeok
18:25florianwhen looking at a profile of startup (eg I&#39;m currently looking at https://perfht.ml/2oZZ0Oy ), is there an easy way to see what code ran before the main browser window was first visible to the user? And what code ran before that window started accepting user events?
18:28mstangeflorian: you can search for the call to show the widget, e.g. on Mac nsCocoaWindow::Show https://perfht.ml/2pN8tvy
18:28mstangethe window becomes visible at the end of this call
18:31florianthanks!
18:31floriando you know which function I should search for on Windows?
18:32florianalso, is it a known bug that for the main process there&#39;s a red bar spanning the whole profile? (event processing delay shown as the length of the whole profile when hovering that bar)
18:36mstangeflorian: nsWindow::Show, I think
18:36mstangeflorian: is that a bug, though?
18:36mstangeooh
18:36mstangeyes it is
18:37florianyeah, it was another way to ask my &quot;And what code ran before that window started accepting user events?&quot; question.
18:37florianwondering if I should file a bug or cc myself to an existing one
18:38mstangeflorian: please file a bug
18:38mstangeI think it&#39;s because we&#39;re starting the profiler before we can even dispatch events to the main thread
18:38mstangewhich only happens with startup profiling
18:40florianmstange: https://bugzilla.mozilla.org/show_bug.cgi?id=1358583
21:33John-Galtmstange: Where should I file a bug about making talos not try to ping the symbolication server on non-Windows platforms?
21:33John-GaltIt slows things down enough in normal cases, but I always wind up getting throttled after a couple of runs, after which things slow to a crawl.
22 Apr 2017
No messages
   
Last message: 123 days and 7 hours ago