On Fri, Oct 2, 2015 at 2:59 PM, Alvaro Herrera <alvherre@2ndquadrant.com> wrote:
> Robert Haas wrote:
>> The standby can have the feature enabled even though the master has it
>> disabled? That seems like it can only lead to heartache.
>
> Can you elaborate?
Sort of. Our rule up until now has always been that the standby is an
exact copy of the master. I suspect deviating from that behavior will
introduce bugs. I suspect having the standby make data changes that
aren't WAL-logged will introduce bugs; not to be unkind, but that
certainly seems like a lesson to take from what happened with
multixacts.
I haven't looked at this code well enough to guess specifically what
will go wrong. But consider people turning the feature on and off
repeatedly on the master, and separately on the standby, combined with
crashes on the standby that restart replay from earlier points
(possibly with settings that have changed in the meantime). Are we
really sure that we're never going to end up with the wrong files, or
inconsistent ones, on the standby? I have a really hard time
believing that's going to work out.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company