450371 - CPU goes 100% busy for a long time on shutdown

Status: RESOLVED WORKSFORME

(SeaMonkey :: General, defect)

Description

[Nelson Bolyard (seldom reads bugmail)]

Running SM trunk nightly build Gecko/2008080600 

Every time I shut down SM, the CPU goes 100% busy for minutes.  
Usually stays that way until I kill it.  

I attached to the process while it was in this state with a debugger. 
I did a little stepping, and found that it is in a tight loop, calling 
NS_ProcessNextEvent_P from nsThread::Shutdown 

See that code at 
http://mxr.mozilla.org/comm-central/source/mozilla/xpcom/threads/nsThread.cpp#463

The stack is:

xpcom_core.dll!nsThread::Shutdown()                  Line 469	C++
necko.dll!nsSocketTransportService::Shutdown()       Line 447	C++
necko.dll!nsIOService::SetOffline(int offline=0x00000001)  Line 622	C++
necko.dll!nsIOService::Observe()                     Line 775	C++
xpcom_core.dll!nsObserverList::NotifyObservers()     Line 128 + 0x12 bytes
xpcom_core.dll!nsObserverService::NotifyObservers()  Line 184	C++
xul.dll!nsXREDirProvider::DoShutdown()               Line 838	C++
xul.dll!ScopedXPCOMStartup::~ScopedXPCOMStartup()    Line 916	C++
xul.dll!XRE_main(	

If I force it to exit that loop and fall through to PR_JoinThread,
the program exits very quickly after that.

Comment 1

[WADA:World Anti-bad-Duping Agency]

> OS: Windows XP

FYI.
Process Monitor can trace file related activity(open/close/read/write etc.) of Seamonkey, and trace of file related activity provides valuable information about "what program is doing" in many cases, not any case though.
> Sysinternals Utilities Index
>  http://technet.microsoft.com/en-us/sysinternals/bb545027.aspx
> Process Monitor v1.37
>  http://technet.microsoft.com/en-us/sysinternals/bb896645.aspx 

Comment 2

[Tony Mechelynck [:tonymec]]

Confirming.

On Linux, I often, but not every time, see seamonkey-bin using between 75% and 95% of the CPU time (with all other processes sharing the rest) after I hit Ctrl-Q and confirm that I want to close all tabs. I don't know what causes it but when it happens, the Sm process doesn't end, even after 10 minutes of inaction on my part. In the ksysguard "System Load" graph (one of the Linux equivalents of the Windows Task Manager) I see an unusually high amount of "system" time being consumed. The only solution I've found so far is to kill it manually with kill -15

Today I'm using

Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1a2pre) Gecko/20080813004850 SeaMonkey/2.0a1pre

but I haven't yet tried to close it since install.

Comment 4

[Nelson Bolyard (seldom reads bugmail)]

I've been seeing this problem on Windows XP A LOT lately.
I've been breaking into the hung process with the debugger and 
looking around, and I've noticed a number of things.  These observations 
have led to a workaround or two, one of which seems satisfactory.

There's always one thread trying to close a socket.  The stack of Windows
system shared libraries on that thread is VERY deep, as if Windows code
called by NSPR to close the socket has called itself recursively a lot.  

On another thread, there's some kind of timer event firing over and over.
Since I don't have sources to match my browser, I cannot tell exactly 
what code is doing that.  I think that one thread is starving the other,
and the starved thread needs to finish for the shut down to complete.

While it is using 100% of the CPU, setting the process priority down from
8 (normal) to 4 (idle) with Windows Process Explorer usually causes it to
stop hanging and terminate normally within a second after doing so.

Going "offline" before exiting the process seems to avoid the hang.  
The process shuts down very quickly when I do that.  I am now doing that
frequently, and have never seen a shutdown hang since I started doing that.

Comment 5

[Nelson Bolyard (seldom reads bugmail)]

I originally reported this using a nightly build from 2008-08-06.
I continued to see it in nightly builds from 
2008-08-16
2008-08-31
2008-09-09
2008-09-17
2008-09-18
2008-09-22
2008-09-30

But since switching to 2008-10-16, this problem seems greatly reduced.
Now, when I shut down the browser, the CPU goes 100% busy for 1-3 seconds,
but no more.  Then the process terminates successfully.  I have not seen
it stay 100% busy for minutes (until it is killed) since switching to that
version.

Comment 6

[Nelson Bolyard (seldom reads bugmail)]

This problem is too different from the symptoms of bug 416745 to be a 
duplicate.

Comment 7

[Wayne Mery (:wsmwk)]

Nelson, what TCP ports are still open when it's in the hung state using CPU?
Is this a regression?

Comment 8

[Nelson Bolyard (seldom reads bugmail)]

I haven't seen this symptom for a long time.  If Tony Mechelynck is also not
seeing it any more, then I think this can be resolved WORKSFORME.

Comment 9

[Roman R.]

Might the cause of this bug be related to the cause of bug 490948?

Comment 10

[Wayne Mery (:wsmwk)]

(In reply to comment #9)
> Might the cause of this bug be related to the cause of bug 490948?

no connection as far as I can see.