Re: 16-bit page checksums for 9.2

On Wed, Feb 29, 2012 at 4:41 PM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Wed, Feb 29, 2012 at 11:24 AM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
>> On Wed, Feb 29, 2012 at 3:46 PM, Heikki Linnakangas
>> <heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
>>
>>>> Are you saying you would accept the patch if we had this?
>>
>>> I think I would still be uncomfortable with the hacks in the page header.
>>
>> There are no "hacks". There are some carefully designed changes with
>> input from multiple people, including yourself, and it copes as
>> gracefully as it can with backwards compatibility requirements.
>
> You have comments from three different people, all experienced
> hackers, disagreeing with this position;

Who is the third person you speak of? Perhaps they will speak again if
they wish to be heard.

> Heikki and I have both
> proposed alternate approaches.

No, you've proposed rejecting the patch in favour of some future
dream. We all agree on the future dream but it clearly happens in the
future and that could easily be more than 1 release ahead.

If you have comments that would allow a patch in this release, I am
happy to hear them. I hear only two people seeking to reject a patch
that adds value for users. Quite frankly, the comments about flag bits
are ridiculous and bogus.

> I'm not sure that we're at a point
> where we can say that we know what the best solution is, but I think
> it is clear that there's enough concern about this that you ought not
> to be denying that there is a problem.

There are some weird cases that can cause problems. My wish to is to
resolve those so everyone is happy. If those weird cases are simply
used as an excuse to reject, then I don't accept that and nor will our
users. Of course, if there was a significant issue, it would be
immediate rejection but there isn't.

I'm trying to commit a useful patch in this release. If you'd like to
work with me to find a solution acceptable to all, I'd be happy to
include suggestions, just as I have already included comments from
Heikki, Bruce, Noah and others. I do accept that Heikki now says that
the code I added at his request is just a hack.

Assuming I'm going to commit something in this release, what should it
look like?

At Heikki's request/idea I will be working on a 2-phase database level
parameter that will give us checksumming on the whole database after a
scan to enable it. That sounds like it will resolve the corner cases
relatively cleanly.

--
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services