Re: High Availability / Clustering

[Dave Augustus <davea@support.kcm.org>]

Hi Lee,

In order to explain this, let us use these definitions:

- the LVS LDAP server- consists of numberous slaves
- the master servers- server A (primary master) and server B (secondary
master)

Assume that the primary master is Active

1.Oops!! 
2.server A goes down!!!
3.Heartbeat restarts server B as the LDAP master and takes over the
4.shared IP address of the master servers.
5.Some time passes.... seconds, minutes,hours, days...
6.server B comes back up as a slave

Because the master LDAP servers don't have many writes, it is entirely
possible that the server A not require any updates when it returns to
service as a slave. If updates happen to server B while server A is
down, a replog.rej is created, as you probably know. 

Therefore, in order to make this 100% automatic, these delayed updates
to server A would have to be added when it returns to service as a
slave. (I have not scripted this part because I just haven't seen any
major problems with the current setup.)

This could be accomplished by running a cron on both master servers that
would try to update its other server if the rej file exists. Of course,
the first thing to determine is whether the local machine is the active
server. If it is, AND if the rej file exists,AND the slave is up, run
the slurpd in one-shot mode on the master server. If it isn't then do
nothing. I currently have a cron that emails me with the latest slurpd
records- I forgot the filename- but anyhow, the email lets me know if I
need to take action.

I would be happy to send you the scripts but you will need LVS, mon and
heartbeat to make this happen.

Dave