On Wed, May 17, 2006 at 03:40:17PM -0400, Patrick Radtke wrote: > when we switch we comment out the sync_server line on the replica to > make sure it won't listen. We have a "as master" config and an "as slave" config - and switch them via symlinks. Similar basic concept though - or at least it will be once everything is actually running... > When it comes to get the master up to speed, then we reverse the roles. Yep - that's the idea. Replicate everything back then switch roles again. I'm going to be interested to see how fast we can effect the switchover. Certainly users will lose their connections. We also have an IP address which migrates with the 'master' status - such that the init script that brings up the master IP address also sets the correct symlinks to ensure that the server is running in master mode. The replication is done to another IP address again - and the init script that brings up that IP also ensures that the server is running in replica mode. So nothing can replicate unless one server is in master mode and the other in replica mode - the IP addresses ensure it. This has the other very nice advantage that the configuration files are _identical_ on the two servers, since we don't have to specify a different replication target. > We also generate a file with all our usernames in it and feed it to > sync_client on the replica and that way it will go through every user > and make sure they get re-synced. Now that's a fantastic idea. Why didn't I think of that. I'll be writing something like that in the next few minutes. I do think you meant to say "sync_client on the master" there. Bron. ---- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
