1183412 - 24 MiB regression in AWSY-e10s RSS after enabling service worker cache

Status: RESOLVED INVALID

(Core :: DOM: Service Workers, defect)

Description

[Eric Rahm [:erahm]]

Attachment: e10s memory report

+++ This bug was initially created as a clone of Bug #1110144 +++

AWSY regression range: https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=d94a96e41f50&tochange=f71f5a88e16d

This was noted on the e10s instance of AWSY [1]. It's measurements are pretty noisy, but this is way above the noise. The e10s instance is a bit behind, so it might be worth waiting a few days to see if things settle down.

I compared w/ non-e10s AWSY and did not see the regression. Unfortunately it's all RSS so I can't get a particularly useful memory diff.

Also note: only the main process regressed, content process seemed unaffected.

[1] While on corporate VPN: http://awsy1.corpdmz.scl3.mozilla.com:8000/index.htm

Comment 1

[Ben Kelly [:bkelly, not reviewing]]

Has the test corpus for AWSY changed recently?  I would be shocked if any of the ancient sites we used to run actually use Cache.  None of the changes in that regression range would matter if Cache is not being used.  If Cache was being used, its memory should show up in heap, sqlite usage, etc.

Also, I can't think of any reason e10s would differ from non-e10s.  They both use IPC via PBackground.

Did anything change in the build infrastructure around that date?

I'm not saying this isn't a problem, but right now I find it hard to believe its Cache.

Comment 2

[Eric Rahm [:erahm]]

(In reply to Ben Kelly [:bkelly] from comment #1)
> Has the test corpus for AWSY changed recently?  I would be shocked if any of
> the ancient sites we used to run actually use Cache.  None of the changes in
> that regression range would matter if Cache is not being used.  If Cache was
> being used, its memory should show up in heap, sqlite usage, etc.
> 
> Also, I can't think of any reason e10s would differ from non-e10s.  They
> both use IPC via PBackground.
> 
> Did anything change in the build infrastructure around that date?
> 
> I'm not saying this isn't a problem, but right now I find it hard to believe
> its Cache.

Nothing changed on AWSY's side. We can wait for a few more runs to come down the pipe to make sure this isn't just a stray report. I can also try to repro locally tomorrow.

Comment 3

[Ben Kelly [:bkelly, not reviewing]]

I think we should look at the test corpus too.  If nothing is executing Cache then the only thing Cache does at startup is what QuotaManager does.  It basically creates a single QuotaClient object for Cache and might iterate over some (non-existent) files.  Cache does not do any aggressive pre-load or pre-caching.

My guess is the size of the binary changed or the binary layout changed in some way.

Comment 4

[Eric Rahm [:erahm]]

Now that there's been some time for the e10s AWSY instance to settle down it looks like this was one time spike. Lets close this for now and if I see again we can reopen.