Database upgrade

Hi all,

We'll do a big database upgrade around 02:00EST Sunday morning. Site will be down for about three hours.

This cluster will be the first db setup with SSD only (Intel DC S3700) and the servers will go from 256 to 768GB memory. This should remove the recent intermittent performance aggravations (for example occasionally slow loading my_uploads pages).

If anyone is interested in details, it's Postgres 9.3 which should now be mature. Old db was 9.1, but there were some new features that were too tempting in 9.3 over 9.2 (better json support, cascading replication, foreign data wrappers etc) so we were not so conservative in the jump this time. From what I understand, the big changes in the binary files where the actual data is stored happened in 9.1->9.2 so 9.1->9.3 should not be more risky. We've run 9.2 and 9.3 on internal test servers for a while.

Best,
Marcus

Thanks for the update

So, the new primary database has a hiccup with the memory. In about 2h we're going to failover to one of the standbys and change the defect memory modules. That will imply a few minutes of downtime coordinating and syncing the drives. Schedule for actual surgery is 02:00UTC.

Sorry about this. Some errors seems to only be triggered with production usage pattern.

Best,
Marcus

We're now on the failover.

Marcus, do you ever go to sleep?

Hope this will also make the statistics work properly again - looking forward to changed memory modules then - thank you Guys for all your Work on the database