murder and kerberos5?

[Patrick H Radtke]

> And another question, can "master and replica" be part
> of a standard murder setup. I'm using
> cyrus-imapd-2.3.1-2. I'll be grateful for any help.


You can use 'master and replica' in a murder environment.
We are running 14 backend pairs in our murder environment.


Since the primary (I call the 'master' machine primary to avoid confusion 
when people associate 'master' with murder master) machine and the replica 
have the same mailbox list, you don't want the replica talking to the 
murder master. We simply comment out the 'mupdate' option from cyrus.conf, 
and the assoicate config options from imapd.conf. That way our replica 
won't talk to the murder master, but if there is failover we can easily 
uncomment out those lines and make the replica active.

Now you have to consider how to fail over in the murder environment.
Failing over from primary to replica means you need to update the mailbox 
list on the murder master and all the frontends. Depending on how many 
mailboxes you have, this can take long time (~20 minutes).

To improve this fail over speed, you can set up a VIF on the primary 
backend. Set the 'servername' option in imapd.conf to the name of that 
vif. The mailbox list on the murder master and frontends would list the 
user mailboxes as being on the VIF server. Then during failover you would 
up the VIF on the replica. You would not need to update the mailbox list, 
since the replica is now serving requests for the VIF. Obviously you 
should down the VIF on the primary machine during the fail over.


Things to watch out for:
Doing an xfer command will move a user between to backends, but the user 
will still exist on the previous machines replica and then also on the 
current machines replica.

If you need to update the mailbox list, make sure the replica is using 
ctl_mboxlist -am . The 'a' will make the replica the 'authoritative' 
source for the mailbox. Otherwise, if you just run ctl_mboxlist -m, the 
replica will delete all your mailboxes! Which is exactly what you don't 
want in a fail over situtaion.

-Patrick