1376904 - [meta] Stop doing full GCs

Status: NEW

(Core :: JavaScript: GC, enhancement, P2)

Description

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

We should stop doing GCs where we have to collect every zone. Bug 1285355 was a blocker for this, but it's fixed now.

Comment 1

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Here is the list of full GCs sorted by frequency. I filed bugs for the first two:

[(u'FULL_GC_TIMER', 2887),
 (u'PAGE_HIDE', 1197),
 (u'USER_INACTIVE', 1134),
 (u'CC_WAITING', 935),
 (u'TOO_MUCH_MALLOC', 355),
 (u'CC_FORCED', 206),
 (u'LOAD_END', 192),
 (u'ALLOC_TRIGGER', 144),
 (u'API', 26),
 (u'MEM_PRESSURE', 17),
 (u'INTER_SLICE_GC', 13),
 (u'SET_DOC_SHELL', 13),
 (u'NSJSCONTEXT_DESTROY', 5),
 (u'COMPARTMENT_REVIVED', 1)]

Comment 2

[Andrew McCreight [:mccr8]]

Note that this includes the case where we only have, say, a couple of zones and we happen to collect them all. For instance, a user with only one tab open may end up with everything as a "full GC".  CC_WAITING, number 4 on the list, never explicitly triggers a full GC.

Comment 3

[Paul Bone [:pbone]]

1376557 will reduce (but probably not remove) ALLOC_TRIGGER non-incremental GCs

Comment 4

[Paul Bone [:pbone]]

Nevermind, I'm wrong.

Comment 5

[Andrew McCreight [:mccr8]]

This bug is about whether we collect every zone in a collection, not whether we are incremental or not.

Comment 6

[Jon Coppeard (:jonco)]

We generally use 'full GC' to mean a GC that collects all zones vs. a 'zone GC' that only collects some zones.