"Serguei Mokhov" <sa_mokho@alcor.concordia.ca> writes:
> Kelly Harmon <kelly.harmon@byers.com> wrote in message news:9pr7f7$k0j$1@news.tht.net...
>> At that point, it seems that your screwed...in that even if you shut down
>> postgres and restart it, somewhere it knows that that database has data that
>> needs to be written to disk, and it refuses to continue until it does so.
> Isn't it the WAL who 'remembers' this info?
Both WAL and pg_log *must* be on writable media, so there's really no
chance of putting the whole of a $PGDATA tree onto a CD. However one
could imagine putting individual databases (or even individual tables)
onto CD. One thing you'd have to watch out for is that Postgres
may try to update on-row commit status bits even during a read-only
operation such as SELECT. The best way to deal with that would be to
VACUUM the table or database before moving it to read-only storage.
VACUUM would leave the status bits all set correctly.
We've talked repeatedly about implementing a notion of tablespaces
to allow DBAs to exercise more control over where tables are kept.
Maybe it'd make sense to allow tablespaces to be marked read-only, too.
regards, tom lane