On Oct 21, 2006, at 6:08 AM, Martijn van Oosterhout wrote:
> On Sat, Oct 21, 2006 at 10:37:51AM +0100, Simon Riggs wrote:
>> Turning off WAL is a difficult topic. Without it you have no crash
>> recovery, which IMHO everybody says they don't care about until they
>> crash, then they realise. It's hard to be selective about writing WAL
>> for specific operations also.
>
> It's been discussed before. One idea is to declare tables without
> logging. The idea being that during recovery those tables and related
> indexes are simply truncated. No foreign keys allowed. Obviously they
> will not be saved via PITR either.
>
> Put another way, the table structure is saved in WAL, but the data
> isn't.
This is exactly what I'd like. Simon suggested turning off WAL
during the loads as a possible hack solution. The reason this won't
work is that we snap all the time, lots of tables. We have between
2000 and 4000 snapshot operations per day (throughout). At the same
time we have reporting queries running (that create and/or populate
other tables) that last from 5 minutes to 18 hours. It is important
that we run everything but the snapshots with WAL on (as we must have
PITR -- sans snapshots)
// Theo Schlossnagle
// CTO -- http://www.omniti.com/~jesus/
// OmniTI Computer Consulting, Inc. -- http://www.omniti.com/