On 16.8.12 15:06, Michael Dürig wrote:
On 16.8.12 14:30, Jukka Zitting wrote:
Hi,
On Wed, Aug 15, 2012 at 10:37 AM, Michael Dürig <[email protected]>
wrote:
No test case but its obvious from the code. There was a condition
variable
earlier which signalled a pending shut down (or
ObservationManager.dispose)
to the ChangeProcessor. The later used that to exit its run method.
I guess we could restore something like that. But note that we can't
control what goes on inside an observation listener registered by a
client, so whatever we do there's no way to guarantee a timely return
from an observation listener. That's why I wouldn't worry too much
about that case, and instead just throw exceptions to any background
threads that are still trying to access a repository that's already
closed.
Throwing exceptions client accesses the repository from an observation
listener which "takes too long" is fine for me. However I think the
repository should stop sending observation events as soon as a shut down
is pending.
I created https://issues.apache.org/jira/browse/OAK-252 to track this.
Michael
Michael
BR,
Jukka Zitting
