Re: postgres_fdw bug in 9.6

From: Jeff Janes <jeff(dot)janes(at)gmail(dot)com>
To: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: Robert Haas <robertmhaas(at)gmail(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: postgres_fdw bug in 9.6
Date: 2016-12-10 18:29:05
Message-ID: CAMkU=1zEAryzjEa-njBkV2+uXXVHHk6u2V2eS1MT+zP7s=vj1Q@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Thu, Dec 8, 2016 at 10:28 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> > Maybe it would help for Jeff to use elog_node_display() to the nodes
> > that are causing the problem - e.g. outerpathkeys and innerpathkeys
> > and best_path->path_mergeclauses, or just best_path - at the point
> > where the error is thrown. That might give us enough information to
> > see what's broken.
>
> I'll be astonished if that's sufficient evidence. We already know that
> the problem is that the input path doesn't claim to be sorted in a way
> that would match the merge clauses, but that doesn't tell us how such
> a path came to be generated (or, if it wasn't intentionally done, where
> the data structure got clobbered later).
>
> It's possible that setting a breakpoint at create_mergejoin_path and
> capturing stack traces for all calls would yield usable insight. But
> there are likely to be lots of calls if this is an 8-way join query,
> and probably only a few are wrong.
>
> I'd much rather have a test case than try to debug this remotely.
> Bandwidth too low.
>

I didn't get an asserts on the assert-enabled build.

I have a test case where I made the fdw connect back to itself, and
stripped out all the objects that I could and still reproduce the case. It
is large, 21MB compressed, 163MB uncompressed, so I am linking it here:

https://drive.google.com/open?id=0Bzqrh1SO9FcEZkpPM0JwUEMzcmc

When running this against and default configured 9.6.1 or 10devel-a73491e,
I get the error.

psql -U postgres -f <(xzcat combined_anon.sql.xz)

...

ERROR: outer pathkeys do not match mergeclauses
STATEMENT: explain update local1 set local19 = jj_join.local19 from
jj_join where local1.local12=jj_join.local12 and local1.local12 in
('ddd','bbb','aaa','abc');

Cheers,

Jeff

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Robert Haas 2016-12-10 18:38:27 Re: Effect of caching hash bucket size while costing
Previous Message Dilip Kumar 2016-12-10 12:59:11 Proposal : Parallel Merge Join