TOPIC:

Hello all!

Thats nice tool, simple and works very fine, We are testing now and I have a question:
Is there any Heartbeat Disk to ensure that hosts are really down?
If network bettewn Pool Hosts fail, but all VM in Pool is running perfectly, What happens?

Thanks
Gunther

Thanks for the positive feedback.

There is no heartbeat disk at this time. We are considering using the XAPI database as a pseudo disk to store and share information between hosts since the next version in development will rely more on the XAPI DB to share information between hosts. The overall concept is to minimize complexity while achieving most of the features of a more typical design which relies on a lot of SW packages and collaboration between them. Our approach is a number of bash scripts working together to achieve the same with no system modifications required.

The design will however provide reasonable assuredness that a host is really down:

Here are the 2 possible scenarios:

1 - Only the Master can fence Slaves. Slaves cannot fence each other. If a Master experiences a network failure and "sees" the Slaves as all being down, the Master is prevented from fencing any slaves and optionally will reboot itself. This is achieved by setting FENCE_REBOOT_LONE_HOST to 1 in the configuration. In this scenario FENCE_BLOCK flag will be set which disables the fencing function.

Please note - if this is the case, chances are a Slave will have already fenced the Master and taken over as master (scenario 2 below)

2 - Only a single dynamically selected Slave is allowed to fence the Master. If the Slave detects a master failure it will need quorum to takeover for the master. If all the hosts are sharing the same network infrastructure, this is a relatively safe approach. Hosts on non-shared infrastructure may be more likely to experience split brain. Additionally, you can set tolerance with FENCE_MIN_HOSTS which will prevent fencing if the set number of hosts are not visible, further minimizing the possibility of split brain.