Re: (ITS#3764) deadlocked bdb

[hyc@symas.com]

richton@nbcs.rutgers.edu wrote:
> Full_Name: Aaron Richton
> Version: 2.2.26+patches
> OS: Solaris 9
> URL: https://www.nbcs.rutgers.edu/~richton/olits/deadlock-20050609.txt
> Submission from: (NULL) (68.239.199.130)
>
>
> I have the "long txn" bdb patches i.e.
> http://www.openldap.org/lists/openldap-devel/200407/msg00068.html on top of OL
> 2.2.26. (I'm interested in DB_LOG_AUTOREMOVE working. Since long txn made it
> into 2.3, I'm interested in that working too!)
>
> The machine I'm using to test the longtxn patch failed to shut down (SIGINT to
> slapd) with some apparent locking issues. The bug details (gdb bt and db_stat)
> are just long enough that I've used the URL field instead of inlining the text.
>
> Since this is only in test, I can leave the server contending for a few days,
> issue a gdb/dbx command if you want, or whatever else might be helpful.
Well, there's an outstanding read lock (locker ID 0x48) but there aren't 
any threads left that could own it. There's only one active thread, 
which is in hdb_modify. It has a pair of transactions active (one is 
nested in the other) and nothing else. It seems that some other reader 
thread exited without cleaning up its read lock. Not sure what else we 
can discover here, because we're not likely to find locker 0x48 in 
anything that's still on any threads' stack.
-- 
  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support