On Tue, Jun 2, 2015 at 11:36 AM, Andres Freund <andres@anarazel.de> wrote:
>> That would be a departure from the behavior of every existing release
>> that includes this code based on, to my knowledge, zero trouble
>> reports.
>
> On the other hand we're now at about bug #5 attributeable to the odd way
> truncation works for multixacts. It's obviously complex and hard to get
> right. It makes it harder to cope with the wrong values left in
> datminxid etc. So I'm still wondering whether fixing this for good isn't
> the better approach.
It may well be. But I think we should do something more surgical
first. Perhaps we can justify the pain and risk of making changes to
the WAL format in the back-branches, but let's not do it in a rush.
If we can get this patch to a state where it undoes the damage
inflicted in 9.3.7/9.4.2, then we will be in a state where we have as
much reliability as we had in 9.3.6 plus the protections against
member-space wraparound added in 9.3.7 - which, like the patch I'm
proposing now, were directly motivated by multiple, independent bug
reports. That seems like a good place to get to. If nothing else, it
will buy us some time to figure out what else we want to do.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company