Richard Huxton wrote:
> It calls archive_command on the just-filled one.
Good to know, thanks. I think I'll experiment a bit with
archive_command. My point was that since I know (or better assume) that
old segments are going to stay in my pg_xlog for *days* before getting
recycled, just copying them all after the call to backup_stop() should
be enough, in my case. It's more than I need, even.
> You do know that pg_dump gives you a guaranteed accurate snapshot of the
> database? It doesn't matter if it's in use.
Yes, I know, that's why I'm using it. A dump is also useful in that you
can restore it even in a different system easier. A text dump allows you
to make changes before restoring, even with sed. I'm comfortable with it.
The reasons for playing with WAL segments for backups are:
1) learning how it works;
2) everything else is backed up with cpio;
3) at restore time I need to have PG running already, and there may be
something different at database initialization.
Reason 1) is the driving one, by far. I can handle it for the rest (i.e.
the compressed dump is saved as part of the cpio archive).
.TM.