Morning all,
I'm not sure if this has been discussed before, but I'm looking at
running the Windows version of the FD within the Windows 2000 cluster
service.
This looks to be possible, with the cluster service handling the
stopping & starting of the process and if the configuration is stored on
a drive which the cluster has control of, then the config can fail over
with the process.
Windows lists the requirements for a cluster-unaware application to be
clustered as follows:
* The application must be able to store its data in a configurable
location.
Check - config file is a command line parameter
The application must use TCP/IP to connect with clients.
Check
Clients must reconnect in the event of intermittent network failure.
This might be a problem - in my experience, a reset of the TCP
connection will cause the job to terminate. There is an option to rerun
failed jobs, but it would be nicer if the director and sd could hold off
terminating the job for a minute or so whilst they try to re-establish
the connection.
If a job is re-run, presumably it is re-run from the beginning, so if we
were doing a backup of 190GiB of data onto a 200GiB tape, and it got to
180GiB and then the cluster node went down, when bacula reruns the job
and connects to the fd again (now running on the backup cluster node)
presumably it will begin the job from the start and overflow the tape
(although I suppose it could rerun it as a differential...).
Has this ever been considered, and how much work would it be to
implement?
--
Russell Howe
russell_howe < at > wr...
