1682319 - [meta] Android killing content process aggressively on some devices (low on memory)

Status: RESOLVED DUPLICATE of bug 1752594

(GeckoView :: Sandboxing, defect, P2)

Description

[Sebastian Kaspari (:sebastian; :pocmo)]

I was debugging the following Fenix issue where users describe that when switching tabs the page always reloads:
https://github.com/mozilla-mobile/fenix/issues/12731

At first I thought our memory trimming code in AC (that suspends tabs) was the culprit. But even when this is disabled I see it happen basically any time I switch tabs on my Nexus 9. Most users in this issue are using a Samsung S8.

Further debugging I noticed that Android calls Application.onTrimMemory() on the Application object of the content process. After giving the final warning with TRIM_MEMORY_COMPLETE Android kills the content process. We keep the state of all tabs and the next time the user switches to a tab we lazily restore it - which visually looks like a reload. Then there are again "too many" tabs and the content process gets killed after multiple warnings and the next switch again looks like a reload.

I opened a PR to further reduce our own memory trimming on the AC side. However that does not affect this killing of the content process and I wonder what we can do in this case?

Comment 1

[(No longer employed by Mozilla) Aaron Klotz]

We need https://github.com/mozilla-mobile/android-components/issues/9366

Comment 2

[Emily Toop (:fluffyemily)]

Any update on progress with this :sebastian?

Comment 3

[Sebastian Kaspari (:sebastian; :pocmo)]

Replied on the GitHub issue. @Aaron: Also feel free to ping me on Slack/Zoom, if that resolves this quicker. :)

Comment 4

[Sebastian Kaspari (:sebastian; :pocmo)]

Heh, just saw in my mails that Emily scheduled a meeting already. 👍

Comment 5

[Jan Henning [:JanH]]

Out of curiosity, does anybody know how Chrome handles this problem? Do they suffer from the same problem, have they found this one weird trick that sidesteps this issue, do they benefit because vendors might commonly whitelist Chrome from their resource usage optimisation routines...?

Comment 6

[Sebastian Kaspari (:sebastian; :pocmo)]

(In reply to Jan Henning [:JanH] from comment #5)

Out of curiosity, does anybody know how Chrome handles this problem? Do they suffer from the same problem, have they found this one weird trick that sidesteps this issue, do they benefit because vendors might commonly whitelist Chrome from their resource usage optimisation routines...?

Not sure how Chrome handles this. But on my Nexus 9 tablet, where this happens on almost every tab switch, I can reproduce this in Chrome too (Tabs reload visually) - but it needs a some of more tabs to be opened before it reproduces.

Comment 7

[Chris Peterson [:cpeterson]]

Tentatively set this bug's priority to P2 so we review it for GV's 2022 H2 planning.

Comment 8

[Chris Peterson [:cpeterson]]

Irene, is this bug actionable? Do your other tab unloading fixes make this bug obsolete or fixed?

Comment 9

[[:owlish] 🦉 PST]

I think it's actionable, and I don't think it's fixed. At least, we need to check if it's still reproduces.

Comment 10

[Chris Peterson [:cpeterson]]

Requires investigation in H2 to see if the tab unloading fixes helped with this issue.

Comment 11

[Chris Peterson [:cpeterson]]

Dupe of bug 1752594

Comment 12

[Chris Peterson [:cpeterson]]

Moving content process management bugs to the new GeckoView::Sandboxing component.