Suppose we have a DB cluster with an additional tablespace and we are
able to make an atomic SAN snapshot of *only* the main cluster
volume...
The additional tablespace contains only UNLOGGED relations.
We cannot snap that volume so we use rsync as follows...
1. pg_start_backup('foo');
make SAN snapshot
rsync the add'l tablespace
pg_stop_backup()
Now provision a new cluster around the snapshot and rsync'd volume,
rejigger the pg_tblspc link if necessary... and start it up maybe or
not having it remain as a streaming replica.
It's been my experience that possibly bulky data in the additional
tablespace does *not* need be rsync'd if we capture only the *_init
files.
Id' be curious to here feedback re the sanity of this approach.
And would also like to know if perhaps *only* the directories under
the rsync'd tablespace actually must be present for a successful
recovery.
The above approach has worked mumerous times even with origin systems
having large, churny contents in the dedicated unlogged tablespace
(which is on a much faster local NVME volume than the main SAN
volume.)
Thanks!