> The only way we can support file-level hot backup is in conjunction with
> PITR-style WAL log archiving. It is okay for the data area dump to be
> inconsistent, so long as your recovery process includes replay of WAL
> starting at some checkpoint before the filesystem dump started, and
> extending to some point after the filesystem dump finished. Replaying
> WAL will correct the inconsistencies.
And the "last checkpoint" info resides in pg_control, and not in pg_clog, no ?
So basically a PITR restore would need to adjust the pg_control file
after filesystem restore and before starting recovery. Maybe it can take that
info from the oldest available WAL ? The OP would only need to ensure,
that only such logs that need to be rolled forward are visible (in the
correct directory) to the recovery.
Andreas