On Tue, Dec 19, 2017 at 10:39 AM, Stephen Frost <sfrost@xxxxxxxxxxx> wrote:
Greetings,
* James Keener (jim@xxxxxxxxxxxxx) wrote:
> Would a storage block level incremental like zfs work?
This really depends on what you want out of your backups and just
exactly how the ZFS filesystem is set up. Remember that any backup of
PG that doesn't use PG's start/stop backup must be atomic across all
tablespaces and even then that really just allows you to bring PG back
up as of that point of the snapshot. I wouldn't recommend trying to
play WAL forward from that kind of a backup. If you use do use
pg_start/stop_backup with ZFS snapshots, and make sure to track all of
the WAL that's generated between the start/stop backup and ensure it's
reliably stored, etc, then they can work, but it's not simple.
I believe that the thread started with a data warehouse use case. That might be one application where data ingestion and processing can be stopped and started in a controlled manner. As opposed to a continuously live system where changes are going to continually accumulate in the WAL.
Best regards,
Alex