RE: deadlock trace file in weblogic environment
Date: Fri, 13 Feb 2009 10:34:04 -0800 (PST)
I find that at least the cvt | held mode part is not right; it's always 5. For instance,
Global Wait-For-Graph(WFG) at ddTS[0.5] :
BLOCKED 0xd933e9a0 5 wq 2 cvtops x1 [0x70011][0x5c8f3],[TX] [34000-0001-0000522D] 0 BLOCKER 0xd933e850 5 wq 1 cvtops x8 [0x70011][0x5c8f3],[TX] [5C000-0001-00000575] 0 BLOCKED 0xd933ec58 5 wq 2 cvtops x1 [0x180017][0x3935f],[TX] [5C000-0001-00000575] 0 BLOCKER 0xd933eaf0 5 wq 1 cvtops x8 [0x180017][0x3935f],[TX] [34000-0001-0000522D] 0
It was the simplest ORA-60 deadlock created on RAC (Oracle 10.2.0.4). Both sessions were in the first instance. Of all RAC deadlock graphs I've seen, mode is always 5. But other pieces of info may be correct.
Hrishy, are you saying you see a lot of lines in alert.log reporting deadlocks but the trace files on filesystem are not found (except one)? If they're removed by somebody or DBA's cron job, just read it from /proc (on most UNIXes/Linux). For instance, on my Linux box, alert.log says /home/oracle/oracle/product/10.2.0/db/admin/riscs1/bdump/riscs11_lmd0_1295.trc has the deadlock trace, but ls says No such file or directory. So I did:
$ ls -l /proc/1295/fd | grep riscs11_lmd0_1295.tr l-wx------ 1 oracle oinstall 64 Feb 13 11:46 2 -> /home/oracle/oracle/archive/riscs1/bdump/20081201/riscs11_lmd0_1295.trc $ vi /proc/1295/fd/2
Obviously, if process 1295 exited, you wouldn't have this luck. But my 1295 is LMD so it won't exit until instance shutdown.
> The wait-for-graph columns are:
> <BLOCKED|BLOCKER> <lockp> <cvt|held mode> <res name> <did|pid|txn_id> <node>
> Vlado Barun, M.Sc.
> Sr. Manager, Database Engineering and Operations
> Jewelry Television
> Mobile: 865 335 7652
> Email: vlado.barun_at_jtv.com
>> There are 100's of messages in the alert log saying a deadlock occured but >> there is only one trace file why is this ? >> The trace files gets appended with the error messages for the dead lock.Received on Fri Feb 13 2009 - 12:34:04 CST