On Mon, Feb 23, 2009 at 2:27 AM, Jan Ancajas <[email protected]> wrote:
> For the backup there shouldn't be a problem, it will only copy the > data to an xml file. During the restore part the unavailability period > occurs from erasing the db up to completely restoring it. The > wait-until-queues-clear-out scenario will not be addressed since we'll > never know how long the pending tasks takes long to finished and we > are restoring the db to a previous state anyway. It seems to me you need to stop accepting new requests and stop building things before making a backup, otherwise you could end up in an inconsistent state. For example, the backup is written, *then* a build finishes, then you go import into a new version, and it's either missing or has the wrong info for that build. This is handled in the current implementation because you have to stop Continuum in order to access the embedded Derby db with the command line tool to do the backup. But it's probably best to take the existing code as-is, and then make improvements. Thanks again! BTW, I didn't see an issue for this, so I opened http://jira.codehaus.org/browse/CONTINUUM-2098 for the Web UI work, (and linked it to -1421 for a way to shut down). -- Wendy
