On 16 October 2014 05:26, Andres Freund <andres@2ndquadrant.com> wrote:
>> I think manual checkpoints should flush everything.
>>
>> This is a valid use case.
>>
>> What other use case is there for a manual checkpoint?
>
> There's people using frequent manual checkpoints to keep performance
> predictable. Unfortunately that actually can improve jitter quite
> measurably.
Hmm, more discussion required there it would seem.
> If we want to change this, fine, but we shouldn't sneak it into the back
> branches, together with a correctness fix
The bug lies in the default behaviour, which we must fix.
I agree backpatching is awkward, though there may also be people who
believe that a CHECKPOINT flushes everything, plus other related bugs
may be lurking.
Seems like we could backpatch this...
CHECKPOINT [ALL (defult) | PERMANENT]
and people who are doing CHECKPOINT PERMANENT for performance reasons
can add the new keyword easy enough, if we highlight it in the release
notes.
--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services