Here is the backtrace of the third svnserve:
#0 0x420e19ee in select () from /lib/i686/libc.so.6
#1 0x400db77c in __DTOR_END__ () from /usr/lib/libdb-4.0.so
BTW, this symbol is never correct (the correct symbol is __os_sleep), I
believe its an error in gdb + optimizations producing this symbol. It
will be different almost every time you recompile a new version.
#2 0x400bd5e5 in __os_yield () from /usr/lib/libdb-4.0.so
#3 0x4005887d in __db_tas_mutex_lock () from /usr/lib/libdb-4.0.so
#4 0x400b3839 in __log_put_int () from /usr/lib/libdb-4.0.so
#5 0x400b33b1 in __log_put () from /usr/lib/libdb-4.0.so
#6 0x400cb938 in __txn_regop_log () from /usr/lib/libdb-4.0.so
#7 0x400ca742 in __txn_commit () from /usr/lib/libdb-4.0.so
#8 0x40033457 in commit_trail (trail=0x8086aa8, fs=0x80795d0)
at subversion/libsvn_fs/trail.c:100
#9 0x4003350d in svn_fs__retry_txn (fs=0x80795d0,
txn_body=0x4003324c txn_body_change_txn_prop, baton=0xbffff0c0,
pool=0x80867a0) at subversion/libsvn_fs/trail.c:136
This looks identical to the hangs I was getting in DB-4.0.14, in either
case I don't think DB-4.1.25 fixed it completely, just made it far more
rare. Most of the time when DB calls __os_yield() its because its
waiting on another thread or process to finish some work. The patch I
posted earlier this week about disabling txn_checkpoint() makes this go
away for all the cases I've seen except for when you overload the
machine with too many simultanous requests (but the server will
eventually handle all the requests correctly, it may take 15+ minutes to
do this though).
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 14 02:06:19 2006