Re: Bugs/slowness inserting and indexing cubes - Mailing list pgsql-hackers

From Alexander Korotkov
Subject Re: Bugs/slowness inserting and indexing cubes
Date
Msg-id CAPpHfduSW5QZGWOE+-Zoe_rQTBTM28xtD3eu9KOyKEuCqWr70Q@mail.gmail.com
Whole thread Raw
In response to Re: Bugs/slowness inserting and indexing cubes  (Alexander Korotkov <aekorotkov@gmail.com>)
Responses Re: Bugs/slowness inserting and indexing cubes  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
ITSM, I found the problem. This piece of code is triggering an error. It assumes each page of corresponding to have initialized buffer. That should be true because we're inserting index tuples from up to down while splits propagate from down to up.

if (!found)
{
/*
* Node buffer should exist at this point. If it didn't exist before,
* the insertion that caused the page to split should've created it.
*/
elog(ERROR, "node buffer of page being split (%u) does not exist",
blocknum);
}

But this assumptions becomes false we turn buffer off in the root page. So, root page can produce pages without initialized buffers when splits.

/*
 * Does specified level have buffers? (Beware of multiple evaluation of
 * arguments.)
 */
#define LEVEL_HAS_BUFFERS(nlevel, gfbb) \
((nlevel) != 0 && (nlevel) % (gfbb)->levelStep == 0 && \
(nlevel) != (gfbb)->rootitem->level)

So, I think we should just do silent return from the function instead of triggering error. Patch is attached.

------
With best regards,
Alexander Korotkov.
Attachment

pgsql-hackers by date:

Previous
From: Bruce Momjian
Date:
Subject: Re: random_page_cost vs seq_page_cost
Next
From: Tom Lane
Date:
Subject: Re: When do we lose column names?