Bitbucket will be unavailable for up to two hours starting Monday, Nov 22st, 02:00 GMT while we upgrade our database to PostgreSQL 9. We expect the downtime not to exceed two hours.
We’ve been looking forward to the new streaming replication feature in PostgreSQL 9 for quite some time and we’re excited about the improvements in reliability this will bring to Bitbucket.
When we migrated Bitbucket from EC2 to Contegix we used Postgres’ WAL archiving feature to minimize downtime while replicating the database. Since then we’ve continued to use WAL archiving to continuously back up our database in addition to twice daily pg_dumps. However, we didn’t have a true hot standby database. If our primary database failed we would have to recover manually and switch our Django settings file to point to the slave database. This all changes with Postgres 9.
We recently deployed two beefy new servers, and have practiced the upgrade in advance. It will take us around 20-30 minutes to do the actual database upgrade, which requires us to take Bitbucket offline.
We’re also taking the opportunity to perform other minor maintenance that will extend the maintenance window, such as reconfiguring our remote access cards, which require server reboots to pick up configuration changes. We’ll try to keep the downtime to a minimum and bring the site back up as soon as possible!
Thanks for your patience as we work to increase Bitbucket’s reliability.