Loading...

derby-user@db.apache.org

[Prev] Thread [Next]  |  [Prev] Date [Next]

Re: Database shutdown not releasing file handles Kristian Waagan Wed Apr 04 03:00:43 2012

On 04.04.12 09:27, Trejkaz wrote:
On Wed, Apr 4, 2012 at 4:46 PM, Kristian Waagan
<[EMAIL PROTECTED]>  wrote:
On 04.04.12 05:23, Trejkaz wrote:

We have the occasional user reporting that after shutting down the
database, they can see file handles still open.

We're using ;shutdown=true (or at least the DataSource equivalent of
it) to do this, and we were led to believe that this would be
sufficient to release all file handles.

Hi,

Yes, it should, but can you please clarify if you're talking about database
shutdown or Derby system shutdown?

Just individual database shutdown.

Also, do you know to what kinds of files Derby keeps open handles?

The files which are held are:

  seg0\c951.dat
  seg0\c861.dat
  seg0\c211.dat
  seg0\c71.dat
  seg0\c130.dat
  seg0\c731.dat
  seg0\cf0.dat

These do seem to be closed at some point *after* shutdown returns,
leading us to believe that shutdown doesn't necessarily shut down
"now", but rather "some time in the future."

Yes, if the handles are closed shortly after shutdown has returned this is unfortunately the currently expected behavior. Not sure if we have a JIRA for that, at least it has been discussed at some point.
Hopefully "some time in the future" is only a few seconds away.


The timing is like this:

  11:09:02.0444360 AM - close call presumably finishes around here
  11:10:00.8759376 AM - we attempt to rename the directory
  11:10:00.8760454 AM - first close call after our rename attempt
  11:10:12.4420715 AM - last close call after our rename attempt

Is it possible that shutdown doesn't actually shut the database down?

Which version of Derby are you running with?

v10.8.2.2.

Another probably-critical piece of information I missed was that the
OS is Windows. (If it had been Linux, we probably wouldn't ever notice
this sort of thing since we discovered the issue when trying to move a
file which was held open.)

Yes, that's typical, and the community sometimes observes the same thing when such a bug is introduced by someone developing on a non-Windows machine and then the nightlies fail on the Windows platform after the commit.


--
Kristian


Daniel