Horton

Re: Re: PANIC: invalid index offnum: 186 when processing BRIN indexes in VACUUM

Alvaro Herrera <alvherre@alvh.no-ip.org>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Michael Paquier <michael.paquier@gmail.com>, PostgreSQL mailing lists <pgsql-hackers@postgresql.org>
Date: 2017-11-01T16:15:47Z
Lists: pgsql-hackers 
Alvaro Herrera wrote:
> Tomas Vondra wrote:
> 
> > FWIW I can reproduce this on 9.5, and I don't even need to run the
> > UPDATE part. That is, INSERT + VACUUM running concurrently is enough to
> > produce broken BRIN indexes :-(
> 
> Hmm, I'm pretty sure we stress-tested brin in pretty much the same way.
> But I see this misbehavior too.  Looking ...

Turns out that this is related to concurrent growth of the table while
the summarization process is scanning -- so new pages have appeared at
the end of the table after the end point has been determined.  It would
be a pain to determine number of blocks for each range, so I'm looking
for a simple way to fix it without imposing so much overhead.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Commits

  1. Fix BRIN summarization concurrent with extension

  2. Fix corner-case errors in brin_doupdate().

  3. Rewrite PageIndexDeleteNoCompact into a form that only deletes 1 tuple.

  4. Invent PageIndexTupleOverwrite, and teach BRIN and GiST to use it.

  5. Close some holes in BRIN page assignment