On Thu, May 3, 2012 at 12:16 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Having said all that, I wasn't really arguing that this was a guaranteed
> safe thing for us to rely on; just pointing out that it's quite likely
> that the issue hasn't been seen in the field because of this type of
> consideration.
Well, we do rely, in numerous places, on writes << 512 bytes not
getting torn. pd_prune_xid, index tuple kills, heap tuple hint bits,
relmapper files, etc. We generally assume, for example, that a 4-byte
write which is 4-byte aligned does not need to be WAL-logged, which
would be necessary if we thought that the write might be torn.
Are you planning to commit Noah's patch?
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company