Compute worker silently fails when XS host runs out of space

[Rick Harris]

This bug only affects compute workers using the XenServer hypervisor.

If a XenServer's Dom0 disk fills up, the compute worker's call to XenServerSession login will hang. This will cause the compute worker service to not start. This will, in turn, cause messages to accumulate on the worker's queue, making it wrongly appear to be a problem in the RPC code-path (Rabbit, carrot, etc).

Proposed fix:

We should be able to detect the login hanging, time it out, and then log an error message on the compute worker. Perhaps something like:

"Unable to login using XenServerSession (is the disk full?)"

[Armando Migliaccio (armando-migliaccio)]

maybe this is related to the bug #692994.

[Sandy Walsh (sandy-walsh)]

Ok, I'm able to reproduce this now. Working on a fix.

[Sandy Walsh (sandy-walsh)]

Hmm, even easier way to reproduce this problem. Just stop the xapi service running on Dom0.

[Sandy Walsh (sandy-walsh)]

Sorry, just stopping the service gives different results (the socket simply can't connect). In low disk conditions login_with_password() completely freezes.