1610228 - [Nightly] refreshing facebook tab seems to continuously increase memory usage for that tab

Status: RESOLVED INCOMPLETE

(Core :: JavaScript: GC, defect, P3)

Description

[Alexander]

Attachment: memory-report.json.gz

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:74.0) Gecko/20100101 Firefox/74.0

Steps to reproduce:

1. log into facebook account
2. after page has loaded reload page by pressing the reload button

Actual results:

looking at firefox in windows 10 task manager keeps showing increasing memory usage.

Expected results:

Memory usage shouldn't increase.

Comment 1

[Alexander]

Attachment: about:support

Comment 2

[Daniel Kolsis]

Hello, I've managed to reproduce the issue on Firefox 72.0.2, Firefox Beta 73.0b8 and 74.0a1 (2020-01-21).

Comment 3

[Daniel Kolsis]

Setting component to Core → Performance, if you feel this is incorrect, please set the correct component.

Comment 4

[Alice0775 White]

I can reproduce the issue on Nightly74.0a1 as well as Firefpx72 Windows10.
In my case,
Good build: Immediately reduce the memory usage to less than 500Mb at every reloads.
Bad build : The memory discarding seems does not work, So, the memory usage more than 900Mb at 10th reloads.

Regression window:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=3a763059a684969df0246d673943ec1d5ff8996b&tochange=5decb743461a478278ee42d9c54c24b939db9d59

Suspect:
5decb743461a478278ee42d9c54c24b939db9d59 Steve Fink — Bug 1593399 - Rework how mark colors are handled in weakmap marking r=jonco

So, this seems to be duplication of Bug 1610193.

Comment 5

[Steve Fink [:sfink] [:s:]]

It does sound very much like bug 1610193, but unfortunately I have a Nightly since that fix and I can reproduce.

Comment 6

[Steve Fink [:sfink] [:s:]]

Hm. Actually, it's a little weird. After reloading facebook 10 times, I was a little over 1GB. I saved CC/GC graph, then went back and remeasured, and it had dropped back to 330MB. So at least post-bug 1610193, it's not a permanent leak. It really seems like it shouldn't build up this much, though. And it would be good to verify this behavior, and compare to an unfixed version.

Comment 7

[Steve Fink [:sfink] [:s:]]

I just reloaded 10 more times and got up to ~650MB -- but 300MB of that was heap-unclassified. After waiting a bit and re-measuring, it again dropped down to 347MB.

Comment 8

[Worcester12345]

Where does one measure this, or how can one help with testing on this bug?

Comment 9

[Steve Fink [:sfink] [:s:]]

I was measuring by going to the about:memory page, doing "Measure", and looking at the total explicit usage for the process that contains the facebook page (which will probably be the first one after the main process; use the navigation links on the top right.) Measure once for a baseline, then do whatever reloading stuff you want and measure again. I was reloading 7 or 8 times in a row, then re-measuring. You could also use the task monitor or equivalent for your OS.