Re: PATCH: 9.5 replication origins fix for logical decoding

On 2015-10-15 17:05:33 +0800, Craig Ringer wrote:
> On 15 October 2015 at 16:48, Andres Freund <andres(at)anarazel(dot)de> wrote:
> > On 2015-10-15 16:02:23 +0800, Craig Ringer wrote:
> >> There's an oversight in replication origins support in logical
> >> decoding, where the origin node ID isn't passed correctly to callbacks
> >> except for the origin filter callback. All other callbacks see it as
> >> InvalidRepOriginId.
> >
> > Only for the transaction, right? I.e. the stuff on changes should be
> > correct? Your description sounds like it's more than that?
> >
> > I don't think your fix is entirely correct, the
> > XLogRecGetOrigin(buf->record) shouldn't be in the XACT_XINFO_HAS_ORIGIN
> > block.
>
> I guess since it'll be InvalidRepOriginId otherwise, that makes sense.

That's not the point. XACT_XINFO_HAS_ORIGIN is about
origin_lsn/timestamp, it doesn't have anything to do with the record
origin (which is included in many more types of record than just
commits).

> > Your test prints the origins from the transaction instead the changes -
> > why?
>
> I don't understand this part.

Your test prints origin in commits - but changes can have individual
origins.

Greetings,

Andres Freund