Closed
Bug 1305091
Opened 9 years ago
Closed 9 years ago
Stop using content processes when they are using "too much" memory
Categories
(Core :: DOM: Content Processes, defect)
Core
DOM: Content Processes
Tracking
()
RESOLVED
FIXED
mozilla54
People
(Reporter: benjamin, Assigned: benjamin)
References
Details
(Whiteboard: [e10s-multi:M3])
Attachments
(1 file)
When a content process starts using "too much" memory, it's much more likely to crash. Depending on why we're using memory, this can also correlate with high GC/CC pause times or other general jank.
So something I think we should do short-term is stop assigning new tabs to an existing content process when it has high memory. We'll call this a "deprecated content process".
More complex (move to another bug/later?) would be to move *existing* tabs to another process on page navigation.
Proposed criteria for marking a content process as deprecated:
* on any system (32-bit/64-bit) >1G private bytes
* on any system: more than one GC pause >500ms in the last hour
* on 32-bit systems: <512M largest available block of VM
Updated•9 years ago
|
tracking-e10s:
--- → ?
![]() |
||
Updated•9 years ago
|
Whiteboard: [e10s-multi:?]
![]() |
||
Updated•9 years ago
|
Updated•9 years ago
|
Whiteboard: [e10s-multi:?] → [e10s-multi:M3]
Comment hidden (mozreview-request) |
Assignee | ||
Updated•9 years ago
|
Assignee: nobody → benjamin
Status: NEW → ASSIGNED
Comment 2•9 years ago
|
||
mozreview-review |
Comment on attachment 8834463 [details]
Bug 1305091 - Stop using content processes when they are running out of memory data-r?francois
https://reviewboard.mozilla.org/r/110398/#review111696
This looks good to me.
Attachment #8834463 -
Flags: review?(ksteuber) → review+
Comment 3•9 years ago
|
||
mozreview-review |
Comment on attachment 8834463 [details]
Bug 1305091 - Stop using content processes when they are running out of memory data-r?francois
https://reviewboard.mozilla.org/r/110398/#review111820
Attachment #8834463 -
Flags: review?(mrbkap) → review+
Comment 4•9 years ago
|
||
mozreview-review |
Comment on attachment 8834463 [details]
Bug 1305091 - Stop using content processes when they are running out of memory data-r?francois
https://reviewboard.mozilla.org/r/110398/#review111876
datareview+
Attachment #8834463 -
Flags: review?(francois) → review+
Pushed by bsmedberg@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/63af206d8770
Stop using content processes when they are running out of memory, r=mrbkap+ksteuber data-r=francois
Comment 6•9 years ago
|
||
bugherder |
Status: ASSIGNED → RESOLVED
Closed: 9 years ago
status-firefox54:
--- → fixed
Resolution: --- → FIXED
Target Milestone: --- → mozilla54
![]() |
||
Comment 7•9 years ago
|
||
(In reply to Benjamin Smedberg [:bsmedberg] from comment #0)
> More complex (move to another bug/later?) would be to move *existing* tabs
> to another process on page navigation.
I talked to Bill about this a while ago and the only realistic way to move tabs/windows between processes currently, would be to reload them (using session store, restore tab probably). Which could be done easily for background tabs, especially since there is also a plan to unload some of the background tabs in some cases. On the plus side that would not be terrible difficult to implement.
>
> Proposed criteria for marking a content process as deprecated:
> * on any system (32-bit/64-bit) >1G private bytes
> * on any system: more than one GC pause >500ms in the last hour
> * on 32-bit systems: <512M largest available block of VM
I guess we will block this functionality on e10s-multi, because of service workers mainly, or do you plan to release it sooner? Do you plan to add a cap for the number of concurrent deprecated processes?
Flags: needinfo?(benjamin)
Assignee | ||
Comment 8•9 years ago
|
||
My current plan was to let this ride the trains now not blocking on e10s-multi, and monitor how frequently it happens. If it happens a lot to certain users (more than 4x/session)s, we may need to implement capping, but right now I doubt that will be a problem.
Flags: needinfo?(benjamin)
You need to log in
before you can comment on or make changes to this bug.
Description
•