Cluster Maintenance in the Wee Hours
We sometimes have to take the cluster down for scheduled maintenance and upgrades. We usually do this at 11:05pm or 12:05am on the Saturday/Sunday changeover. We did this this weekend.
We did a number of upgrades and improvements, including:
-
Replaced two older SAN units with new ones. This will increase disk I/O performance for a number of customers.
-
Updated firmware in all SAN units.
-
Re-sync’d a database that had gotten out of sync with its replica on Friday evening.
-
Increased temp and heap table space, which should reduce disk I/O, and increased innodb_buffer_cache as well, potentially easing disk I/O. The combination should result in better database performance.
-
Upgraded early cluster servers to same kernel and slightly newer Xen to be the same as the newest servers in the cluster.
Getting ready at the data center 2 hours before the maintenance window.
Working on one of the servers.
Swapping SAN units.
Upgrading software and firmware.
Ready to restart the cluster.
Share your thoughts with @engineyard on Twitter