[Linux-HA] Heartbeat + asterisk / failback
Adam Moffett
adam at plexicomm.net
Wed Oct 5 08:37:08 MDT 2005
>
> If you've created a CRM (2.0-style) cluster, then you can set the
> failback to occur at a certain time - by setting a time-based policy
> for failing back at that time (like between 2 and 3 AM).
>
> We don't currently have a policy which lets you decide exactly when to
> fail back yourself. But, you could write a program which detects this
> condition, and modifies the policy, then waits a while, and changes
> the failback policy back to "don't fail back".
>
> Andrew could give you the details on how to do both of these options.
>
>
Asterisk has a built in function to stop when there are no callers (they
call it "stop when convenient"). Given a symmetric cluster, is it
possible for the primary node to pick back up when it detects a failure
in the backup?
The sequence would be something like this:
1. node 1 fails
2. node 2 takes over
3. node 1 has come back, so node 2 issues "stop when convenient"
4. node 2 fails (because there are no calls, and it has stopped)
5. node 1 takes over
Feasible?
If that's not feasible, you're saying I need to write a program which
can monitor the condition of the nodes and modify the cib.xml file when
appropriate. I can probably do that, but I was trying to (K)eep (I)t
(S)imple (S)tupid. I guess the time based policy would fit the simple
qualification, but it sounds dangerous. People pretty much expect their
phones to just work, and if their phone isn't working they get all bent
out of shape about it. So the fewer interruptions the better.
The other option would be to not care which machine is running the
resource. The only reason we want to rely on node1 is that node2 is
using rsync to copy it's configs from node1. Can I run an arbitrary
program on a failure? I.e. when node1 fails, node2 stops running
running rsync and node1 starts running rsync when it comes back. So now
node2 is primary until such time as it fails. Is that type of operation
feasible? An active/passive where we don't care which one is active?
More information about the Linux-HA
mailing list