Re: Re: xReader, double-effort (was: Temporary tables under hot standby)

On 5/3/12 2:54 PM, Josh Berkus wrote:
>> (2) If logical transactions had been implemented as additions to
>> > the WAL stream, and Slony was using that, do you think they would
>> > still have been usable for this recovery?
> Quite possibly not.

The key advantage that I see in londiste/slony replication is that your data stream has absolutely nothing to do with anything binary or internal to Postgres. That means that the only way corruption will travel from a master to a slave is if the corruption is in the actual fields being updated, and even that's not a given (ie: UPDATING a field to a completely new value would not propagate corruption even if the old value of the field was corrupted).

So, embedding a logical stream into WAL is not inherently bad... what would be bad is if that "logical" stream was susceptible to corruption due to something like full page writes. Simply embedding the exact same info slony or londiste captures into the WAL should be fine (though likely defeats the purpose). Translating binary WAL data into DML statements would very likely allow corruption to travel from master to slave.
--
Jim C. Nasby, Database Architect jim(at)nasby(dot)net
512.569.9461 (cell) http://jim.nasby.net