Re: BDB 4.4 + forking svnserve weirdness (issue 2564)

On 6/7/06, Garrett Rooney <rooneg@electricjellyfish.net> wrote:

> Well, I just added some printfs to the calls that open and close the
> bdb environment, and it does seem like we've got one more open than we
> do close in the case where there's a failure. The trick is figuring
> out why between the global pool destruction and the bdb close code
> something gets lost...

Ok, so I figured that part out. The issue is pool
creation/destruction ordering. The bdb environment cache is created
after the connection pool, so it's destroyed before it (Last In, First
Out). This means that by the time apr_terminate comes around and
destroys the connection pool the bdb environment has already been
paniced as part of the cache destruction process.

Note that this is only an issue in the test cases, since we run with
--bdb-txn-nosync, normally this stuff would all have been safely
written out to disk. Regardless, it would be awfully nice if the
tests, well, worked, so I'll probably just commit the pool destruction
change.

-garrett

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Wed Jun 7 20:58:33 2006