[CentOS] X session lockup (CentOS 4.7)

Sun Mar 29 15:57:39 UTC 2009
Bart Schaefer <barton.schaefer at gmail.com>

I just had a very odd incident with my session locking up, so I
thought I'd describe it to see if it rings any bells with anyone.

Background:  I shut down my machine last night for Earth Hour, so the
last thing I did before logging out was exit Firefox3, telling it to
save my open tab state.  When I turned the system back on and started
FF3 again, it installed new versions of three plugins (I believe
NoScript and Elasticfox but I don't recall the third) and then
restored the session as expected.

Not related to FF, but I also started up the "filelight" filesystem
visualization app, which I don't normally have running, and it was
still up on a different desktop while all the following was happening.

This morning I opened a new FF3 window to browse CNN.  I opened tabs
on five or six stories from the front page and then went to read
through them, closing each tab as I finished each story.  (There were
six other tabs open in the first FF window, so no more than 12-14 tabs
in two windows total.)  After closing the second such tab FF began
responding very slowly, even to simply scrolling the window, and
eventually stopped responding entirely (wouldn't even redraw when I
dragged windows around).  I popped up an xterm and ran "strace" on FF
to see if I could find the problem, and lo and behold as long as
strace was running FF recovered and responded fine, but when I stopped
tracing it froze again.  Unfortunately I lost the trace output because
of what happened next, but I do recall that it was accessing
/etc/localtime repeatedly and very frequently.

I left strace running and went back to FF, hoping to see the tail of
the trace at the moment FF locked up again.  However, the next time FF
appeared to be frozen I found that not just FF but the entire Gnome
session was stuck.  I was able to move the mouse and to Alt-F2 into a
text console window, but nothing else responded.  I went back and
forth a few times between the text console and the X console
attempting to diagnose (including killing the strace and sending a
CONT signal to FF, because "ps" said FF was in stopped ("T") state),
but after a few passes at that even Alt-F2 stopped working.

Fortunately I have another computer here, so I ssh'd in and ran
"killall -1 firefox-bin".  This woke things up to the point that one
of the pending Alt-F2 keystrokes was seen and the text console came
up, but although I could then switch consoles again, nothing else in
the X session was responding.  I finally HUP'd gnome-session (which
stopped everything else; I checked "ps" to make sure nothing that had
been associated with that session was still running) and I have now
logged in again and everything seems fine.  FF restarted and restored
the session from the time I killed it, and an strace of it now doesn't
show any accesses to /etc/localtime at all.

I find this all especially strange because the whole system had been
restarted less than 12 hours before.