"John Moore" <NOSPAMnews@NOSPAMtinyvital.com> writes:
> In the case of PostgreSQL, as far as I can tell, one could lose all data
> since the previous dump if one lost the database media. In Oracle or
> Informix, that is *not* true, because they can do a point-in-time restore
> from the last full save, based on the WAL's.
If you are archiving the WAL logs, then in theory you could recover
from those in Postgres as well. In practice, I consider this argument
irrelevant, because no one is going to want to work that way.
(Nigh-infinite offline storage for the logs, plus huge recovery time if
you do suffer a crash ... I don't think so.)
A more reasonable approach to getting better-than-hardware reliability
is replicated servers. We have some crude ways of replicating data now,
and should have much better ways in a release or two. (See
http://www.greatbridge.org/genpage?replication_top for some info on
stuff that will likely get rolled into the standard distribution
eventually. I consider Postgres-R the most promising approach.)
As of today, I wouldn't try to run an airline reservation system on
Postgres either. But check back in a year or so.
regards, tom lane