Re: replica catchup to master

[Igor Brezac <igor@ipass.net>]

On Tue, 19 Nov 2002, Quanah Gibson-Mount wrote:

> Hello,
>
> I was wondering if anyone has some input on how you would get a new replica
> to catch up to the master.  Basically, assume that you are adding a new
> replica to the pool.  Due to the large volume of changes you incur, you
> can't stop slurpd on the master and then do a slapcat, install the DB on
> the replica, and then turn slurpd back on.  In Netscape directory server,
> it keeps track what changes a replica has received by a change number, so
> if you load a new replica with an older DB, it catches it up.  I don't see
> the functionality for this in OpenLDAP.  Therefore, assuming I load a new
> replica with an older DB, after adding it in as a replica in the slapd.conf
> file on the master, how do I get the master to replay changes to the
> replica?
>

Here is what I did:

1. Reconfigure slapd.conf for a new replica
2. Stop slapd and slurp
3. Take a snapshot of the ldap db filesystem (man fssnap)
4. Start slapd/slurp
5. Copy (standard filesystem copy) ldap db to the replica host
6. Start replica ldap server
7. unsnap the ldap db filesystem (fssnap -d).

Your downtime is no longer than 5 to 10 seconds.

You do not have to rush 5, 6 and 7.  fssnap is a Solaris 8 and 9 feature,
Veritas file system has had this feature for quite some time.  I believe
Linux supports snapshot as well.

Hope this helps.

-Igor