Andreas Pflug <pgadmin@pse-consulting.de> writes:
>> In principle you could do this today, but we don't have enough
>> support code in place to make it work smoothly, eg WAL segment files
>> aren't labeled with enough identifying information to let you manage
>> an archive full of 'em. Still it doesn't seem that far away.
> So I issue CHECKPOINT, and tar the cluster or database. Still, I got two
> questions:
> - how to restore a single database
You don't. As I said, any physical backup is going to be
all-or-nothing. These techniques are not a replacement for pg_dump.
> - while tar is running, CHECKPOINT(n+1) might be recorded in some files,
> while others have CHECKPOINT(n). How does the backend know to rollback
> to CHECKPOINT(n)?
That's part of the management code that we need to write before this
will really be very useful; you need to be able to associate the
starting time of a tar dump with the most recent previous CHECKPOINT
in the WAL logs. Not hard in principle, just takes some code we ain't
got yet.
regards, tom lane
