[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Yet another replication problem.

I second using sync-repl. I've been using it for about a year in production and it's worlds better than slurpd.

When I was using slurpd for replication it would always get out of sync. Slurpd really isn't suited for wan links because it expects that there will never be a communication failure between Master and Slave. If the link goes down your out of sync until to manually resync, which is a chore with slurpd. Sync-repl is much more durable and resilient. With sync-repl it uses time stamps on the entries to know if an entry should be modified or not. This means that if your Slave looses connection with the Master it will automatically resync once the connection is reestablished. I wrote a howto in the wiki a while back on how to configure this behaviour.

Administration of slaves is much easier with sync-repl. I remember with slurpd I would have to put the Master in readonly so that I could export an ldif that was completely uptodate to import into the slave. With sync-repl you can use any old ldif, or none at all, and it will automatically sync it back up.

Cheers, James.

Jose Ildefonso Camargo Tolosa wrote:

On 5/22/05, matthew sporleder <msporleder@gmail.com> wrote:

Switching to sync-based replication may be the easiest way.

Somebody told me that sync-repl has some problems, can somebody confirm about this?

Thanks for your answer.


On 5/19/05, Jose Ildefonso Camargo Tolosa <ildefonso.camargo@gmail.com> wrote:


I made a post several moths ago, but I still haven't solved the
problem.  I have a 25 replicas scenario, 5 of them are connected via
ethernet (directly), 20 of them are connected by frame relay

I'm using openldap 2.2.23 and tried 2.2.24 (compiled from source) with
berkley db 4.2 (Official Debian Sarge packages).  I'm using 2.2.23

I'm using slurp replication, and about 20 out of the 25 replicas work
just fine, but with 5 of them (one of wich is located in the local
network) I have "rejects", all of the rejects are related to a missing
previous replication (ie, at some point the directories gets
"out-of-sync", and from that time on, you get rejects like: No such
attribute (something like that) or no such object).  I know that 3 of
these sites has power failures wich are unavoidable (and thus can
cause problems).

Does anybody have a deployment this big working?, I'm running out of
ideas (yes, I have a good cache size in the bdb (about 160Mb, the
computer has 2Gb, and the replicas has a 30Mb cache and 256Mb RAM)).
The directory is about 4000 objects in size.  I'm using samba with
ldap backend.

Any ideas,

Thanks in advance,


Ildefonso Camargo
icamargo at merkurio dot com dot ve
icamargo AT unet dot edu dot ve
ildefonso_camargo AT yahoo dot com
ildefonso dot camargo at gmail dot com