651319 - Firefox/Aurora/Nightly hangs on shutdown with high CPU if open for a long time

Status: RESOLVED INCOMPLETE

(Firefox :: General, defect)

Description

[Gordon P. Hemsley [:GPHemsley]]

I don't have precise STR or any useful information, but LegNeato asked me to file this, so here goes.

I'm on Mac OS X 10.6 (though I think this also would occur on 10.4). I tend to leave my Firefox/Aurora/Nightly instances open for a long time. (Basically for days, until they hang or crash or I decide it's time to update.) When I do decide to shut down (or, more likely, restart), Firefox tends to hang until I give up and just force-quit it. Most of the time, I'm restarting to install an update, but I don't know if that's related. AFAIK, I don't really have this problem if I've only had Firefox open for a few minutes (like if I have to restart multiple times because I just updated to an old update). I also tend to have a lot of tabs open, often into the hundreds.

I don't know if that helps any, but that's about all I know about it. This has been going on for a while (many versions), so I've just come to live with it—it's not like I restart a lot, anyway. But if we could track down the cause and fix it, that'd be awesome, too.

Comment 1

[Kai Liu]

As someone who keeps his browser running for weeks on end with hundreds of tabs (the specter of restoring that many tabs is why I never restart), I personally know that this has been an issue that's been around since... forever.  IMHO, the problem is the teardown; freeing memory from a heap is never cheap.  Especially if much of that was paged out.

See:
Bug 407981 (blamed on large Places DB)
Bug 444915
Bug 474699 (suspected JS, probably heap-related)
Bug 474980 (ditto)
Bug 625875 (blamed on heap teardown)

So, which bug to dupe to?

In any case, this is a complicated issue...

Comment 2

[Gordon P. Hemsley [:GPHemsley]]

(In reply to comment #1)
> As someone who keeps his browser running for weeks on end with hundreds of tabs
> (the specter of restoring that many tabs is why I never restart), I personally
> know that this has been an issue that's been around since... forever.  IMHO,
> the problem is the teardown; freeing memory from a heap is never cheap. 
> Especially if much of that was paged out.

Glad to know I'm not the only one. And I concur with the "forever".

> See:
> Bug 407981 (blamed on large Places DB)
> Bug 444915
> Bug 474699 (suspected JS, probably heap-related)
> Bug 474980 (ditto)
> Bug 625875 (blamed on heap teardown)
> 
> So, which bug to dupe to?

Hard to tell. At least three of them seem like likely candidates (if they're not already dupes of each other). But, for the record, I do not clear history on exit, and I don't have Ubiquity installed (at least, not with this profile).

> In any case, this is a complicated issue...

Agreed.

Comment 3

[Wayne Mery (:wsmwk)]

Gordon, is this still happening?  And is CPU high, zero, or in between?

Comment 4

[Gordon P. Hemsley [:GPHemsley]]

(In reply to Wayne Mery (:wsmwk) from comment #3)
> Gordon, is this still happening?

I think so. On the rare times where I actually close the application, I usually use Force Quit because it seems to be taking too long.

> And is CPU high, zero, or in between?

As I recall, CPU is high.

Comment 5

[Gordon P. Hemsley [:GPHemsley]]

I accidentally let the shutdown process go to completion just now (with Aurora 12), and, well, it didn't complete. Rather, it showed a dialog box at the end that was mostly off screen and that, AFAICT, did not have actual content. It did, however, have OK (I think) and Cancel buttons. The OK button just sat there throbbing, and clicking on either button did nothing, as far as I could see.

I eventually wound up force-quitting anyway, though I don't think the CPU was too high at the time. (I admit I didn't really look at it, though.)

Comment 6

[Wayne Mery (:wsmwk)]

can you get a stacktrace?

Comment 7

[Gordon P. Hemsley [:GPHemsley]]

This still happens, even if only open for a few days. (Though I find that the wait at shutdown is proportional to the amount of time the browser is left open.) I've now tried to keep the open time to a few days max, in part to avoid this issue.

(In reply to Wayne Mery (:wsmwk) from comment #6)
> can you get a stacktrace?

I don't think I know how.

Comment 8

[Wayne Mery (:wsmwk)]

(In reply to Gordon P. Hemsley [:gphemsley] from comment #7)
> (In reply to Wayne Mery (:wsmwk) from comment #6)
> > can you get a stacktrace?
> 
> I don't think I know how.

use Activity Monitor, select the Thunderbird process and click "Sample Process", cut and paste the results into a file, attach the file to this bug report please

Comment 9

[Henrik Skupin [:whimboo][⌚️UTC+1]]

I noticed the same or a similar problem on my Linux box a couple of times now. It happens when I shutdown Firefox after using it for a while. Most times it was no longer than 1h. But so far I'm not able to reproduce the behavior constantly. It happens rarely.

I wonder how to get a good stack for it. I tried with gdb but when trying to connect to the process gdb failed. Any proposals what to do when I hit the bug again?

Comment 10

[Olli Pettay [:smaug][[e-mail]]]

Use a profiler to see where the time is spent. sysprof is rather simple but should work just fine
on Linux.

Comment 11

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Attachment: profile.txt

I have seen this bug again after a long time. So this time I did a profiler run with sysprof as Olli suggested. Attached you can find the profiler information. As it looks like the high cpu load in my case is caused by liboil. Not sure what this is, but maybe Olli can help us here.

Comment 12

[Olli Pettay [:smaug][[e-mail]]]

Is this still valid?