TOPIC:

I unplugged management network from master again this morning, and slave did not take over services.

You'll see in the master log that link was lost Dec 16 09:32:45. Slave doesn't seem to notice and continues to report that master is fine.

Thanks for your help.

Log file does not appear to be attached. Can you make sure ha is enabled for the pool?

"ha-cfg status"

Uploaded now. HA is enabled.

Can you describe your network setup? I notice there is an additional ip "172.16.14.47" on the same subnet as the management 172.16.14.199

You are correct, the slave continues to see the master as live while its management link is down. If the slave were unable to communicate with the master we would see calls to XAPI time out on the slave, which is not happening.

You can try to further isolate this by running a test outside of halizard.

Pull the MGT link on the master and then from the slave try running an xe command like "xe host-list". With no access to the master, this should hang on the slave. Based on the logs, I would expect this to work. Maybe there is some other network path back to the master.

Maybe also look at your ARP table in case 172.16.14.47 and 172.16.14.199 are bound to the same MAC on more than 1 physical interface on the master.

Thanks, that command did not hang and I could still ping master from slave. Xen was advertising that IP on the other interface.

Removed secondary IP and bonded the two connections instead. Failover now working well.

Only one problem left: DRBD not starting automatically after reboot on both hosts.

I've attached iscsi-ha filtered log.

As soon as I run "drbdadm up iscsi1" everything is fine.

Looks like DRBD started and is running, but the resource is not loaded.

Is there anything out of the ordinary with the backing device? Possibly a SW RAID that is slow to start?

Can you grab a snippet from dmesg that includes the host booting and starting DRBD for the first time and post here.
thanks