Red Hat Bugzilla – Bug 193482
xmms i386 build hangs in buildsys
Last modified: 2007-04-18 13:43:42 EDT
Something makes the xmms build hang in the i386 devel builder, x86_64 and ppc
are fine. I cannot reproduce that locally in mach or mock. I'm intentionally
leaving the currently stuck job there (currently in progress for 9 hours) so
someone can possibly debug it.
See also bug 193460
In fact, except for arts-devel umask problem, I can rebuild xmms by myself.
For now, I have no idea for xmms building hang.
Looks like this is indeed a problem with i386/devel only, FC-5 built fine.
Created attachment 130224 [details]
Diff to check where building of xmms hangs
Ville, the diff file I attached is VERY UGLY, but I think this is useful for
detecting where buildsys hangs for rebuilding xmms.
Ville -- is this still happening? (of course, now the devel tree is busted today)
Some more data points from debugging this last week:
1) Only happens building i386 jobs on an x86_64 build host
2) pstree output shows the last child process is 'sed', spawned by libtool. No
idea why sed is hanging
3) strace on the sed process shows its sitting in futex lock or whatever that is
4) gdb is useless here for backtraces; apparently after FC3/RHEL4 gdb got a
better ability to show symbols for 32-bit binaries with a 64-bit gdb, but we're
screwed right now
5) pstack shows sed is sitting in the vDSO somewhere around 0xffffe405 (iirc)
So either I need to start building custom gdbs, or I need to figure out some
other way to determine WTF the hung processes are doing. Blocked pipes? I have
Hangs in futex tend to be "fun with the kernel" -- unfortunately, I've never
gotten a better way of really debugging them :-/
Have we tried rebooting the builder?
Not recently. Want to try it on hammer1?
It's been kicked
Seeing that the kerry package which was reported to have the same problem was
built successfully today made me try it again, but unfortunately the xmms build
still hangs (job 9945). Cc'ing Warren, IIRC you were trying to debug this some
The kernel upgrades seem to have fixed this, thanks.