778285 - Failure to free compartments for closed tabs causes horrible performance near OOM

Status: RESOLVED INCOMPLETE

(Firefox :: General, defect)

Description

[Wesley W. Garland]

Attachment: about:memory output

So, I'm working on an app and accidentally allocated about 1 GB of JS objects.

My machine goes into swap, the browser becomes very unresponsive, and I close the offending tab  (only one open besides a text file).

The compartment is *still* taking up memory, even several minutes later as I type this bug, making my browsing experience simply awful. A beach ball on my mac every few seconds, long pauses.

From about:memory, I ran the CC and GC before snapshoting the first attachment. I forget if I ran "minimize memory" or not.

More than 10 minutes later, I still have 1GB of memory allocated to that compartment, and that tab is most definitely closed.  That's when I snapshotted the second about:memory and about:compartments.

Comment 1

[Wesley W. Garland]

Attachment: about:memory a few minutes later

Comment 2

[Wesley W. Garland]

Attachment: about:compartments

There is definitely no tab open to any chiptic.wes.msgs.net content

Comment 3

[Wesley W. Garland]

I followed these directions https://wiki.mozilla.org/Performance:Leak_Tools#Cycle_collector_heap_dump

The file is 385 megabytes uncompressed, about a 42M download, caveat transforor

http://www.page.ca/~wes/cc-edges-1.36851.log.gz

Comment 4

[Andrew McCreight [:mccr8]]

There are some ghost windows, even.

Comment 5

[Justin Lebar (not reading bugmail)]

We do give the JS engine a hint to maybe run a GC when a window is closed, but I don't know the particulars of how the JS engine decides when to actually run a GC.

Comment 6

[Andrew McCreight [:mccr8]]

I think it always ends up running a GC 5 seconds after a page close.  Plus Wes was manually making it GC.

Comment 7

[Justin Lebar (not reading bugmail)]

(In reply to Andrew McCreight [:mccr8] from comment #6)
> I think it always ends up running a GC 5 seconds after a page close.  Plus
> Wes was manually making it GC.

Oh, indeed.

Wes, can you reproduce this problem?  Can you reproduce it in safe mode?  What version of Firefox are you running?

Comment 8

[Nicholas Nethercote [inactive]]

MemShrink:P3 for now -- Wes, can you answer the questions?  Thanks!

Comment 9

[Wayne Mery (:wsmwk)]

(In reply to Justin Lebar [:jlebar] from comment #7)
> (In reply to Andrew McCreight [:mccr8] from comment #6)
> > I think it always ends up running a GC 5 seconds after a page close.  Plus
> > Wes was manually making it GC.
> 
> Oh, indeed.
> 
> Wes, can you reproduce this problem?  Can you reproduce it in safe mode? 
> What version of Firefox are you running?

Comment 10

[Wesley W. Garland]

Hi -- sorry for my bugzilla absence, I have had a stupid-busy schedule since mid-2012 and no longer work after hours due to RSI. Wayne, thanks for the e-mail tickle.

I have not reproduced that issue recently, but I also have not been working on that application in quite some time.  That said: I am scheduled to resume work on that application in about 10-12 days.  I will to try reproduce with Firefox.current then.

Comment 11

[Wayne Mery (:wsmwk)]

thanks

Comment 12

[Wayne Mery (:wsmwk)]

don't think wes is coming back, so clearing NI.

worth keeping open?