New issue
Advanced search Search tips
Note: Color blocks (like █ or █) mean that a user may not be available. Tooltip shows the reason.

Issue 729008 link

Starred by 1 user
Status: Assigned
Owner:
hirosh...@chromium.org
Cc:
perezju@chromium.org
skyos...@chromium.org
altimin@chromium.org
delph...@chromium.org
█ benhenry@chromium.org
primiano@chromium.org
bashi@chromium.org
Components:
EstimatedDays: ----
NextAction: ----
OS: Android
Pri: 3
Type: Bug

Blocking:
issue 624697



Sign in to add a comment

Address memory regressions of "remove EventSender from ImageLoader"

Project Member Reported by perezju@chromium.org, Jun 2 2017 
Splitting off from  issue 728081 .

A bisect found r475528 the suspect of a memory regression.

Among the many alerts caught at around that revusion:
https://chromeperf.appspot.com/group_report?rev=475528

there are indeed several improvements as well as regressions.

Looking at the high level metrics in:
https://chrome-health.googleplex.com/health-plan

the high level overall metrics look fine, so we're probably OK with respect to system health plan.

Still, many of the larger regressions appear to be from measurements while Chrome is in the background, and there are no corresponding improvements for those.

I would suggest focusing e.g. on:
https://chromeperf.appspot.com/report?sid=73d61ca98f66698e324e3f5053ead60b41ef2945bf04420ba150fce1d272518f

A large 20 MiB regression on cc:effective_size while in background.

Perhaps there are some caches or something that were being cleared before the CL and are not anymore?

Comment 1 by hirosh...@chromium.org, Jun 2 2017

Cc: delph...@chromium.org primiano@chromium.org
Components: Blink>Loader
> I would suggest focusing e.g. on:
> https://chromeperf.appspot.com/report?sid=73d61ca98f66698e324e3f5053ead60b41ef2945bf04420ba150fce1d272518f
Thanks for suggestion!
According to the graph, there was another similar regression ( Issue 711044 ) that is fixed by simply reverting the suspected change.

https://bugs.chromium.org/p/chromium/issues/detail?id=711044#c23
contains an analysis of the regression cause -- "has pushed out the onLoad event by several seconds", thus looks like there seems a common root cause.

> many of the larger regressions appear to be from measurements while Chrome is in the background

Hmm. My change uses a DOMManipulation TaskRunner instead of the global timer task runner. I expect the DOMManipulation TaskRunner is associated to a frame and can be throttled while it is in the background, while the timer task runner used previously is not. This might be the cause of larger regressions in the background.

Comment 2 by delph...@chromium.org, Jun 6 2017

Cc: skyos...@chromium.org

Comment 3 by delph...@chromium.org, Jun 6 2017

Cc: altimin@chromium.org
►

Sign in to add a comment