Horton

Re: Combination of ordered-set aggregate function terminates JDBC connection on PostgreSQL 9.6.5

Tom Lane <tgl@sss.pgh.pa.us>

View thread
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Heikki Linnakangas <hlinnaka@iki.fi>
Cc: David Rowley <david.rowley@2ndquadrant.com>, Lukas Eder <lukas.eder@gmail.com>, PostgreSQL mailing lists <pgsql-bugs@postgresql.org>
Date: 2017-10-12T16:34:52Z
Lists: pgsql-bugs

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Treat aggregate direct arguments as per-agg data not per-trans data.

  2. Allow the built-in ordered-set aggregates to share transition state.

  3. Fix AggGetAggref() so it won't lie to aggregate final functions.

  4. Prevent sharing transition states between ordered-set aggregates.

Attachments

I wrote:
> Thinking about it more clearly, if a transition function is being run
> on behalf of several different Aggrefs (with identical input states),
> then no, the transition function should not care which Aggref it
> looks at.  As you say it mustn't do anything different on the basis of
> the finalfn-related fields.  The problem occurs when a finalfn calls
> AggGetAggref --- then, I think that the finalfn is entirely entitled
> to expect that it will see its own Aggref, not some other one that
> happens to share input+transition.

Concretely, I think we need to do the attached.  This seems like
a bug fix to me, so I'm inclined to back-patch it.  In the back
branches we could put the extra AggState field at the end, to
minimize ABI-break hazards.

BTW ... I was quite surprised to notice that the aggdirectargs
are treated as a property that has to be matched in order to
combine transition states.  They aren't used during the transition
phase, so this seems like a pointless constraint.  We could move
the aggdirectargs ExprState list to AggStatePerAggData and treat
the aggdirectargs as part of the finalfn-related data, instead.
As long as we're not merging states at all for OSAs, this is
moot, but it seems like something to fix along with that.

			regards, tom lane