TOPIC:

Hi,

I am trying your project on 2 nodes XCP1.6 pool, and it is working great in normal condition soft shutdown, restart, maintenance mode.

But when we try hard shutdown/reboot in order to simulate hardware fault we ran into some problems.
The VMs under HA watch where properly transfered to the remaining host, but the hard rebooted slave was evicted from the pool.

The slave was not recognised as member of the pool anymore and we had "no interface configured" (we coul sill ssh on it).

When we xe commands the answer was the "Host was deleted from the master's database".

We had to erase the /var/xapi/state.db and then reintroduce the server to the pull. We though that it was caused by FENCE_HOST_FORGET=1 so we cahnge it to 0. And reproduce the scenario, nothing.

Thus I wonder wether I did something wrong or wether it is a mis-configuration.

Thanks for your help.
Exept that your product is just really great.

As of version 1.6.41 it is no longer mandatory to set FENCE_FORGET_HOST to 1. This used to be required to ensure that XenServer would not attempt to start a VM on a failed host. Should a host become forgotten, we include a script in /etc/ha-lizard/scripts named recover_forgotten_slave. This essentially does the same thing you described - erase local DB, reset netwroking, etc..

So, in answer to your question, nothing was done wrong.. Moving forward you can safely set FENCE_FORGET_HOST to 0 to avoid forceful removal of a host. We will update the current documentation to better reflect this change.

Also, for 2-node pools we just released an addon module for HA-Lizard which allows you to have a 2-node pool which is local storage backed (no need for external SAN). This may be of interest to you if working with small pools.

It does Work know with fencing disable. It even works great.

Concerning the ISCSI system it is already in place on our production site, thus we will continue using.

Great job guys, your piece of software is wonderfull.