The db1 upgrade is going to happen soon. I'd like to do it next monday (one week from today). I'll be scheduling an outage for Monday night. I'd like for everyone who can be around, to be around at that time. Also keep a special eye out the next day for any oddities. The basic premise is this:
1) Turn off postgres and mysql on db1 2) Run a dump of both db's 3) import both db's to db2 4) test 5) point all of our applications at db2 6) Wait a week or two then rebuild db1 7) Schedule another outage, move the postgres db back to db1.
We'll let it run of db2 for a week or two then rebuild db1. Once both boxes are at the same package level we'll migrate the postgres db back to db1. Both databases will have mysql and postgres installed but db1 will be running postgres and db2 will be running MySQL. Each will have a regular job to dump and copy their db to the other. This gives us dedicated resources as well as a quick recovery in the event of some catastrophic failure of one of them.
Any questions?
-Mike
Any questions?
Any reason not to use MySQL replication instead of doing the dump between boxes (I don't know Postgres, but I assume it has something similar)?
Just wanting the resources to be dedicated to each DB platform or use it more as a backup instead of a hot failover?
Just curious... not really much functional difference.
Ray
Ray Van Dolson wrote:
Any questions?
Any reason not to use MySQL replication instead of doing the dump between boxes (I don't know Postgres, but I assume it has something similar)?
Just wanting the resources to be dedicated to each DB platform or use it more as a backup instead of a hot failover?
Just curious... not really much functional difference.
The big difference was memory consumption and time. I'm not against setting this up but there's other stuff going on so I don't have time to do it (not to say that one of the other sysadmin-main guys couldn't do it)
-Mike
On Mon, 2007-07-02 at 08:38 -0700, Ray Van Dolson wrote:
Any questions?
Any reason not to use MySQL replication instead of doing the dump between boxes (I don't know Postgres, but I assume it has something similar)?
Just wanting the resources to be dedicated to each DB platform or use it more as a backup instead of a hot failover?
Aside from the resource issues (running two dbs on the same box is competing for the same resources) I've done this before and it was a bit harder than it should be. Mysql has a tendency to stop replicating between master and slave. Then you either have to put the server into read-only and rsync or interpret the mysql errors and back out/repair the offending mysql statements/tables.
Postgres was more stable as a warm-standby but the server was less loaded. (terminology is slightly different because with postgres, you are performing a continuous recovery so you can't use the standby as a read-only replica while it is pulling the replica data.) But it was much more effort to setup. Also, this requires postgres 8.2 and I think we're only going to upgrade to 8.1.
-Toshio
infrastructure@lists.fedoraproject.org