[Linux-HA] what to do on loss of network

[alex at crackpot.org]

Quoting Steve Wray <steve.wray at cwa.co.nz>:

<snip>
> I tested what I have produced and it *kind* of works. Ie:
>
> On the complete failure of network connectivity on the node which was
> secondary at time of failure the following sequence takes place:
>
> 1. The secondary node notices loss of connectivity to the primary.
> 2. It becomes primary.
> 3. It notices loss of connectivity to the network as a whole and
> reverts to secondary again.
>
> When networking is restored it works better than before and required no
> manual intervention (previously drbd had to be told what to do by an
> operator).

Steve :

I want to implement something identical : 'loss of all ping nodes'  
means 'go secondary and stay that way until you have a live ping  
node'.  With cluster nodes in remote data centers, it's the only way I  
can think of to prevent a split-brain.

I think I will have some time next week to try experimenting with what  
you've done so far.  Just wondering if you've made any new discoveries  
since your last post.

thanks,
alex