Hourly LVM snapshots of the production databases
Right now we only generate an LVM snapshot once every 24 hours, which is used to populate our staging databases. This has proven to be inadequate as a recovery technique. Instead we need to create these snapshots at least once an hour, and keep them around for at least 24 hours; 48 if possible. This would allow us to quickly restore disk snapshots without having to use (potentially messy) Azure tools, restore databases using PostgreSQL restoration tools, or something else.
@jnijhof: Since you're our LVM expert (as far as I know) I'm assigning this to you, feel free to re-assign if necessary.