On 2016-05-04 18:22:27 -0400, Robert Haas wrote:
> On Wed, May 4, 2016 at 6:06 PM, Andres Freund <andres@anarazel.de> wrote:
> >> Some of the proposals involve fairly small tweaks to call
> >> MaintainOldSnapshotTimeMapping() from elsewhere or only when
> >> something changes (like crossing a minute boundary or seeing that a
> >> new TransactionId has been assigned). If you can disentangle those
> >> ideas, it might not look so bad.
> >
> > Yea, if we can do that, I'm ok. I'm doubtful about releasing with the
> > current state, even leaving performance aside, since fixing this will
> > result in somewhat changed semantics, and I'm doubtful about significant
> > development at this point of the release process. If it comes down to
> > either one of those I'm clearly in favor of the latter.
>
> How would the semantics change?
Right now the time for computing the snapshot is relevant, if
maintenance of xids is moved, it'll likely be tied to the time xids are
assigned. That seems perfectly alright, but it'll change behaviour.
> So, I was worried about this, too. But I think there is an
> overwhelming consensus on pgsql-release that getting a beta out early
> trumps all, and that if that results in somewhat more post-beta1
> change than we've traditionally had, so be it.
*If* that's the policy - cool! I just don't want to see the issue
not being fixed due to only wanting conservative changes. And the
discussion around fixing spinlock related issues in the patch certainly
made me think the RMT aimed to be conservative.
Greetings,
Andres Freund