(Illustration by Gaich Muramatsu)
On Fri, Jun 25, 2004 at 03:09:03PM +0200, Gabriel Wicke wrote: > Crash during bonnie testing: > > gdb where > #0 0x401822a6 in sigsuspend () from /lib/tls/libc.so.6 > #1 0x080c8738 in SigChoke (sig=357693596) at sighand.cc:241 > #2 <signal handler called> > #3 0x0807509e in fsobj::Open (this=0x75b67a48, writep=0, execp=0, > truncp=0, cp=0x15521e18, uid=0) at fso.h:686 I've seen this backtrace before. It is unusual in the way that it looks like it crashed in the 'IsDir' test. It looks like the object got pulled out from under us even though we still had a reference count on it. I guess your local cache held fewer files than were being used in the bonnie dataset so there was replacement going on and some destructor must be a bit too destructive. The other crash you reported on the wiki (disconnected startup crash) is something I hadn't seen before. It is somewhat clear what is going on. Right after startup the fake /coda/ directory must be still empty and instead of returning ENOENT the directory handling code decided that this is unusual enough to trigger an assertion, I'll try to replace it with returning ENOENT back up to the caller. The only thing that puzzles me is why this doesn't happen every time we (re)start venus. JanReceived on 2004-07-01 15:56:49