Hannu Krosing wrote:
>Andreas Pflug kirjutas K, 26.11.2003 kell 12:09:
>
>
>>ow wrote:
>>
>>
>>
>>>It appears there's not a lot of interest in discussing the possibility of FK
>>>constraint creation WITHOUT the verification check. How then should one handle
>>>the situation with pg_restore and large dbs where creation of FK constraint(s)
>>>may take hours?
>>>
>>>
>>>
>>>
>>I'd prefer a backup/restore method that dumps physical data, so at
>>restore time there's no need for recreation of FKs. But I didn't see any
>>feedback on this proposal either.
>>
>>
>
>Was this proposal a separate one from using WAL logs for PITR ?
>
>
Yes, I mentioned it just a few days when discussing dependency in pg_dump.
This is somewhat complementary to WAL and PITR. I'm seeking for a fast
way to dump and restore a complete database, like physical file copy,
without shutting down the backend. I was thinking of a BACKUP command
that streams out the files including any indexes and non-vacuumed
tuples. A database recreated from that wouldn't be as clean as a
pg_dump/pg_restored database, but it would be up much faster, and there
wouldn't be any dependency problem.
This doesn't really replace pg_dump/pg_restore, because it probably
wouldn't be able to upgrade a cluster. Still, it would be helpful for
disaster recovery.
Regards,
Andreas
