452489 - clean up nsMemoryImpl for OOM issues.

Status: RESOLVED INVALID

(Core :: XPCOM, defect)

Description

[Doug Turner (:dougt)]

Attachment: patch v.1 w/ API change

Changes:

1) adjusts the behavior of the heapMinimize method to make the "immediate" parameter unused.  If heapMinimized is called on the main thread, it will happen synchronously, if it is called on any other thread, it will be called asynchronously.  This is an optional change, but it does make the API easier to understand.

2) I am enabling the memoryflusher thread.  Since we were not holding a reference to the timer, it would never fire (the timer's destructor kills the timer).  I also removed a bunch of member variables are are just not used.

3) I reduced the timer callback from a huge number to once every 1/2 second.  I can't imagine this timer being useful if not called often.

4) On the OSSO platform, we do not use their API but instead look at the sys vm files instead.  This was a suggestion made by one of the maemo developers and allows us to use a higher water mark (we get notified sooner of OOM).


I think we might also want to disable this flusher on windows.  It isn't being used and i do not think it will make any difference.

Comment 1

[Doug Turner (:dougt)]

Attachment: same patch without api changes

lets do it this way first...  no api changes.

Comment 2

[Doug Turner (:dougt)]

Attachment: removes the flusher timer and fixes osso low memory predicate

Comment 3

[Doug Turner (:dougt)]

changeset:   18562:6b6054736cb2
tag:         tip
user:        Doug Turner <[e-mail]>
date:        Mon Sep 01 17:40:15 2008 -0700
summary:     bug 452489 - removes the flusher timer and fixes osso low memory predicate.  also fixes OSSO oom predicate. r=bsmedberg

Comment 4

[Doug Turner (:dougt)]

backed out.  this light up the windows unit test:

http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1220316401.1220322388.28758.gz&fulltext=1

Something that wasn't tested on the try servers. :-(

Comment 5

[Doug Turner (:dougt)]

the box is still red after the backout.  this can land again when the tree clears.

Comment 6

[Doug Turner (:dougt)]

[e-mail]
Wed Sep 03 16:06:30 2008 -0700	04a28fb0458b	Doug Turner — Bug 452489 - clean up nsMemoryImpl for OOM issues. r=bsmedberg

Comment 7

[Doug Turner (:dougt)]

same problem as comment #4.  backed out.

Comment 8

[Doug Turner (:dougt)]

I pushed only the OSSO part:
1799243b9089	Doug Turner — Bug 452489 - Fixes OOM predicate on OSSO to use the high water mark. r=bsmedberg default tip 

The rest of this is going to be blocked by the TestTimer.exe crash on win32 (bug 453822)

Comment 9

[Doug Turner (:dougt)]

Attachment: same as the last patch w/o the OSSO changes (r=bsmedberg)

Comment 10

[Doug Turner (:dougt)]

not going to look at this for a bit;  feel free to reassign.

Comment 11

[Doug Turner (:dougt)]

This is now invalid.  Bug 475595 addresses all of the concerns here and has landed.