1686100 - Route Service.onTrimMemory notifications through to Gecko's memory pressure observer

Status: RESOLVED FIXED

(GeckoView :: General, enhancement, P1)

Description

[(No longer employed by Mozilla) Aaron Klotz]

Since Android may send these to the services hosting content processes, we should route them through to the memory-pressure observer, which will help by forcing GCs, cleaning up the heap, etc.

I don't expect this to be a definitive fix for bug 1682319, but it should help.

Comment 1

[(No longer employed by Mozilla) Aaron Klotz]

Attachment: Bug 1686100: Route Service.onTrimMemory notifications through to Gecko's memory-pressure observer; r=#geckoview-reviewers

Gecko already broadcasts memory-pressure events from the parent process down
to its content processes, but that excludes any case where Android is calling
Service.onTrimMemory directly on the content processes.

This patch handles onTrimMemory and translates its requests into
memory-pressure events that various Gecko components (including the GC)
listen for. We only handle the severity levels for background processes for now,
as memory cleanup for background processes is currently the most pressing issue
(and I'm not even sure what we could do for foreground memory cleanup without
janking anything).

The argument to memory-pressure should depend on the severity of the request
as well as the frequency of pressure events. In particular:

• The most severe request, TRIM_MEMORY_COMPLETE, is a last-ditch effort to
  reduce memory. This should trigger a full GC, so we use low-memory here.
• "New" requests, where "new" is defined as an event that was triggered in
  excess 10 seconds since the previous "new" event, also gets a full GC, so we
  use low-memory here as well.
• Additional requests that do not fall under either of the above criteria
  receive the low-memory-ongoing event, which omits a full GC. The idea here
  is that, when low memory notifications occur in rapid succession, repeatedly
  doing full GCs over and over is just going to hang the Gecko thread and not
  result in any notable improvement.

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/cddfeeb87193
Route Service.onTrimMemory notifications through to Gecko's memory-pressure observer; r=geckoview-reviewers,agi

Comment 3

[(No longer employed by Mozilla) Aaron Klotz]

Attachment: Bug 1686100: Follow-up to fix lint failure; r=bustage

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1dc8d0dd5e30
Follow-up to fix lint failure; r=bustage

Comment 5

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/cddfeeb87193
https://hg.mozilla.org/mozilla-central/rev/1dc8d0dd5e30