Relax...
The past few days have been spent stressing about, planning, and repairing some database corruption. But that's all done now.
And I'm learning the mysql binlog format, which I'll need to know soon since I'm scheming to make a MySQL replication star hub, where this hub in the middle replicates from all servers, splits/merges queries based on users into other binlogs, and all servers replicate back from it. The end effect will be all LJ users on 2 or 3 random hosts, with one sometimes "hot" (last modified) until replication is caught up and they're all marked in-sync, then the load balancing can be applied when a new handle is needed. (This is a variation on the MySQL proxy idea I had earlier, but without the need to do all those extra network copies....) Also, if the hub dies or needs to be restarted, nothing breaks. The "get a user's DBs all in sync" process just gets behind and the load balancing is fixed until things are caught back up with the hub restarted. Fair bit of hand-waving still. May happen, may not.... still working out details. MogileFS is getting mature, though, so I have high hopes for the same concepts applied to MySQL.
Anyway, fuck all that. It's the weekend and it's time to finally relax. We had a party/BBQ here last night and had some beers and played croquet, but I was still working throughout the party, sneaking away from time to time to work on database corruption.
Today I plan to bike around Hagg Lake and read my new Postfix book. I'm finally gonna learn mail so I don't hate it! At least, that's the plan.
And I'm learning the mysql binlog format, which I'll need to know soon since I'm scheming to make a MySQL replication star hub, where this hub in the middle replicates from all servers, splits/merges queries based on users into other binlogs, and all servers replicate back from it. The end effect will be all LJ users on 2 or 3 random hosts, with one sometimes "hot" (last modified) until replication is caught up and they're all marked in-sync, then the load balancing can be applied when a new handle is needed. (This is a variation on the MySQL proxy idea I had earlier, but without the need to do all those extra network copies....) Also, if the hub dies or needs to be restarted, nothing breaks. The "get a user's DBs all in sync" process just gets behind and the load balancing is fixed until things are caught back up with the hub restarted. Fair bit of hand-waving still. May happen, may not.... still working out details. MogileFS is getting mature, though, so I have high hopes for the same concepts applied to MySQL.
Anyway, fuck all that. It's the weekend and it's time to finally relax. We had a party/BBQ here last night and had some beers and played croquet, but I was still working throughout the party, sneaking away from time to time to work on database corruption.
Today I plan to bike around Hagg Lake and read my new Postfix book. I'm finally gonna learn mail so I don't hate it! At least, that's the plan.