Closed
Bug 1036970
Opened 10 years ago
Closed 10 years ago
2.14% linux32 tp5 private bytes regression on inbound (fx33) seen on July 9 from rev fc4d6b5ee8ce
Categories
(Core :: DOM: Workers, defect)
Tracking
()
RESOLVED
WONTFIX
People
(Reporter: jmaher, Assigned: bkelly)
References
Details
(Keywords: perf, regression, Whiteboard: [talos_regression])
Here we have graph server showing a bump: http://graphs.mozilla.org/graph.html#tests=[[257,131,33]]&sel=1404857051000,1405029851000 I did some retriggers to reduce any doubts: https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=f3b37ad2b42f&tochange=07520461d3aa&jobname=Ubuntu%20HW%2012.04%20mozilla-inbound%20talos%20tp5o in general we went from a value of 480MB -> 490MB Here is a link to the changeset: https://hg.mozilla.org/integration/mozilla-inbound/rev/fc4d6b5ee8ce I don't think this is showing up on other platforms, although in due time we will know. Is there anything simple to help reduce this?
There's nothing platform-specific going on there so it doesn't make sense to me that only linux would change...
Assignee | ||
Comment 2•10 years ago
|
||
How does Talos know when to measure? Perhaps we are keeping the new tab tile worker alive a bit longer and its racing with the measurement.
Assignee | ||
Updated•10 years ago
|
Assignee: nobody → bkelly
Status: NEW → ASSIGNED
Reporter | ||
Comment 3•10 years ago
|
||
here is how we get the private bytes: http://hg.mozilla.org/build/talos/file/c36273c3498d/talos/cmanager_linux.py#l75 we have a loop to collect counters here: hg.mozilla.org/build/talos/file/c36273c3498d/talos/ttest.py resolution is set to 20 (i.e. 20 ms). This is done independent of any events on the browser, we start collecting, fire up the browser, test, close the browser, stop collecting.
Assignee | ||
Comment 4•10 years ago
|
||
It seems windows had a similar regression at the same time, so probably not linux only.
Assignee | ||
Comment 5•10 years ago
|
||
I did took three memory reports before and after the PBackground change right after launching firefox. I made sure to run about:newtab in each case so that the background tile worker would run and in turn load PBackground. Across the three runs I saw in two cases that adding PBackground reduced memory by 0.8MB. In one case memory increased by 8MB. In general, though, this just seems like normal start noise. I see that the talos repo is here: http://hg.mozilla.org/build/talos But it does not include tp5, so I can't run that test. If I can get access to tp5 I'll try taking a memory reports after executing the test. Joel, can you point me to tp5?
Flags: needinfo?(jmaher)
Assignee | ||
Comment 6•10 years ago
|
||
Also, lets see what AWSY says once it catches up to July 9. Currently the last data point is on July 7. https://areweslimyet.com/
Assignee | ||
Comment 7•10 years ago
|
||
Got tp5... Thanks Joel! Will try to figure out how to get memory reports from it.
Flags: needinfo?(jmaher)
Assignee | ||
Comment 8•10 years ago
|
||
AWSY has caught up to July 9 and I don't see a corresponding memory increase.
Assignee | ||
Comment 9•10 years ago
|
||
I've been digging into the AWSY reports a bit more. I do see a slight bump in heap unclassified for this rev, but its well within the noise range for the tests. My current theory is that the PBackground changes are slightly slowing shutdown (since we have to synchronously close the actor on the worker) which is changing the timing of the measurement for talos. Joel, based on the AWSY results I'm inclined to mark this WONTFIX. What do you think?
Flags: needinfo?(jmaher)
Reporter | ||
Comment 10•10 years ago
|
||
this makes sense. Thanks for taking a bit of time to look into this and help come up with an idea of what is going on.
Status: ASSIGNED → RESOLVED
Closed: 10 years ago
Flags: needinfo?(jmaher)
Resolution: --- → WONTFIX
You need to log in
before you can comment on or make changes to this bug.
Description
•