Brad Fitzpatrick (bradfitz) wrote in lj_maintenance,
Brad Fitzpatrick

green, sub 6

If you're on Green, subcluster 6 (Where am I?), your journal will be read-only for the next 15 minutes.

During this time we're making a snapshot (slowly, to not kill the machine) which we'll then do integrity checks on before moving all the Green-6 users to another database.

We're in the process of moving all users around onto new hardware, w/ new storage systems, and new database types. This is a big pain, but it'll be worth it.

For the curious: we're moving from lots of small MyISAM databases on 32-bit hardware and mix of master-master and master-slave, to non-partitioned (on a single machine) databases w/ InnoDB tables on 64-bit hardware w/ lots of memory, all master-master, also sitting on LVM2 for quick snapshots. A transitional step is moving all users to InnoDB on 32-bit hardware since we're still stress-testing MySQL on 64-bit and finding some issues which is looking into. Overall, though, we're seeing incredible results and concurrency from InnoDB. The disks are busier than ever and aren't being artificially starved. There was a deadlock issue in 4.0.19-4.0.21, but 4.0.18 works great, as should 4.0.22 (where they reverted to the old locking behavior).

Comments for this post were disabled by the author