Re: Old row version in hot chain become visible after a freeze

On Tue, Sep 5, 2017 at 9:44 PM, Michael Paquier
<michael(dot)paquier(at)gmail(dot)com> wrote:
> As the portion doing vacuum freeze is the one blowing up, I think that
> it is possible to extract an isolation test and include it in the
> patch with repro.sql as the initialization phase.

Indeed, I have been able to reduce that to an isolation test which
crashes with a rate of 100% if assertions are enabled.

I have also spent a couple more hours looking at the proposed patch
and eye-balling the surrounding code, and my suggestion about
heap_tuple_needs_freeze() is proving to be wrong. So I am arriving at
the conclusion that your patch is taking the right approach to skip
freezing completely if the tuple is not to be removed yet if it is for
vacuum either DEAD or RECENTLY_DEAD.

I am adding as well in CC Álvaro and Andres, who reworked tuple
freezing in 3b97e682 a couple of years back for 9.3. Could you look
more at the bug and the attached patch? It does not fundamentally
change from what has been proposed first, just did the following:
- Updated set of comments that incorrectly assumed that only DEAD
tuples should have freeze skipped.
- Added isolation test for the failure.

An entry has been added in the next open CF to track this problem.
This should not fall into the cracks!
--
Michael