1357810 - Investigate source of memory_pressure event in gecko, and whether to repurpose it

Status: NEW

(Core :: JavaScript: GC, enhancement, P3)

Description

[Kannan Vijayan [:djvj]]

Gecko reports a memory_pressure event which initiates GC.  This was supposedly introduced due to requirements by B2G, and may not be relevant any more.

Can we use a free memory metric and repurpose this event to make it relevant again?

Comment 1

[Eric Rahm [:erahm]]

I thought linux, windows, and osx all had this concept too?

Comment 2

[Eric Rahm [:erahm]]

I guess there's prior art in bug 664291.

Comment 3

[Mike Conley (:mconley) (:⚙️)]

Found at least one place where it's fired:

http://searchfox.org/mozilla-central/rev/7aa21f3b531ddee90a353215bd86e97d6974e25b/xpcom/base/nsMemoryImpl.cpp#115-149

Comment 4

[Eric Rahm [:erahm]]

(In reply to Mike Conley (:mconley) from comment #3)
> Found at least one place where it's fired:
> 
> http://searchfox.org/mozilla-central/rev/
> 7aa21f3b531ddee90a353215bd86e97d6974e25b/xpcom/base/nsMemoryImpl.cpp#115-149

I think that's only triggered by nsIMemory::heapMinimize, which is used in imglib [1] and NPAPI's NPN_memflush [2].

[1] http://searchfox.org/mozilla-central/rev/7aa21f3b531ddee90a353215bd86e97d6974e25b/image/ImageFactory.cpp#225
[2] http://searchfox.org/mozilla-central/rev/7aa21f3b531ddee90a353215bd86e97d6974e25b/dom/plugins/base/nsNPAPIPlugin.cpp#926

Comment 5

[Mike Conley (:mconley) (:⚙️)]

Setting keyword perf:responsiveness because this is about GC timing, and GC blocks the main thread.