On Tue, Nov 8, 2016 at 6:52 PM, Martin Sivak <msivak@redhat.com> wrote:

Hi,

mom-vdsm.service contains:

Requires=vdsmd.service After=vdsmd.service

After does not mean much, since vdsm is not integrated with systemd. Systemd does not wait until vdsm is ready to receive connections. We can use systemd.daemon to notify systemd that vdsm is ready, making dependencies more reliable. See https://github.com/oVirt/ovirt-imageio/blob/master/daemon/ovirt_imageio_daem...

So when Shira restarted vdsm, mom was also restarted.

[journalctl --unit vdsmd] Nov 08 18:25:27 RHEL7.2Server systemd[1]: Stopping Virtual Desktop Server Manager... Nov 08 18:25:27 RHEL7.2Server vdsmd_init_common.sh[3053]: vdsm: Running run_final_hooks Nov 08 18:25:27 RHEL7.2Server systemd[1]: Starting Virtual Desktop Server Manager...

[journalctl --unit mom-vdsm] Nov 08 18:17:23 RHEL7.2Server systemd[1]: Starting MOM instance configured for VDSM purposes... Nov 08 18:25:16 RHEL7.2Server systemd[1]: Stopping MOM instance configured for VDSM purposes... Nov 08 18:25:29 RHEL7.2Server systemd[1]: Started MOM instance configured for VDSM purposes.

But mom then immediately failed with:

2016-11-08 18:25:08,008 - mom.RPCServer - INFO - ping() 2016-11-08 18:25:08,010 - mom.RPCServer - INFO - getStatistics() 2016-11-08 18:25:17,028 - mom.RPCServer - INFO - RPC Server ending 2016-11-08 18:25:24,705 - mom.GuestManager - INFO - Guest Manager ending 2016-11-08 18:25:26,575 - mom.HostMonitor - INFO - Host Monitor ending

2016-11-08 18:25:29,869 - mom - INFO - MOM starting 2016-11-08 18:25:29,905 - mom.HostMonitor - INFO - Host Monitor starting 2016-11-08 18:25:29,905 - mom - INFO - hypervisor interface vdsmjsonrpcbulk 2016-11-08 18:25:30,029 - mom.vdsmInterface - ERROR - Cannot connect to VDSM! [Errno 111] Connection refused 2016-11-08 18:25:30,030 - mom - ERROR - Failed to initialize MOM threads Traceback (most recent call last): File "/usr/lib/python2.7/site-packages/mom/__init__.py", line 29, in run hypervisor_iface = self.get_hypervisor_interface() File "/usr/lib/python2.7/site-packages/mom/__init__.py", line 217, in get_hypervisor_interface return module.instance(self.config) File "/usr/lib/python2.7/site-packages/mom/HypervisorInterfaces/vdsmjsonrpcbulkInterface.py", line 47, in instance return JsonRpcVdsmBulkInterface() File "/usr/lib/python2.7/site-packages/mom/HypervisorInterfaces/vdsmjsonrpcbulkInterface.py", line 29, in __init__ super(JsonRpcVdsmBulkInterface, self).__init__() File "/usr/lib/python2.7/site-packages/mom/HypervisorInterfaces/vdsmjsonrpcInterface.py", line 43, in __init__ .orRaise(RuntimeError, 'No connection to VDSM.') File "/usr/lib/python2.7/site-packages/mom/optional.py", line 28, in orRaise raise exception(*args, **kwargs) RuntimeError: No connection to VDSM.

The question here is, how much time does VDSM need to allow jsonrpc to connect and request a ping and list of VMs?

Even if we fix vdsm, mom (and any other client) should be more robust and do not created so much noise when connection is refused. Client should retry connection and log less dramatic warnings. Adding devel@ovirt, vdsm-devel mailing was abandoned more then 2 years ago. Nir