pg.ddx.io  pgsql-admin@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: Teodor Sigaev <teodor@sigaev.ru>
To: Craig McElroy <craig.mcelroy@contegix.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>
Cc: PostgreSQL Admin List <pgsql-admin@postgresql.org>
Subject: Re: Problem with PITR Past Particular WAL File
Date: Fri, 26 Oct 2007 16:53:09 +0400
Message-ID: <4721E335.8090601@sigaev.ru> (raw)
In-Reply-To: <4E5ADC27-53C9-4A63-8BE2-7563E2EC89EB@contegix.com>
References: <FF045AC2-D3E0-4DC6-BE39-4925AE9AD804@contegix.com>
	<10123.1193250394@sss.pgh.pa.us>
	<1FAEADD9-FE4A-444E-9B45-D6192B0265E5@contegix.com>
	<16979.1193327974@sss.pgh.pa.us>
	<4720DAF9.6080002@sigaev.ru>
	<4E5ADC27-53C9-4A63-8BE2-7563E2EC89EB@contegix.com>



Craig McElroy wrote:
> Teodor:
>   Thanks for the followup.  We will be upgrading this system to 8.2.5 in 
> the next day or so and will test it on that anyways just to be sure it 
> isn't related.  Let me know if there is anything else we can provide to 
> help track down the bug.  We will be keeping this base backup and WAL 
> files around so that we can use them for testing.
If I'm right then 8.2.5 will not help you. Attached patch should fix your 
problem, but I suppose that there is an another bug. Test specially developed 
for reproduce your problem discovers another issue. The diagnostic is:
FATAL:  failed to add item to index page in 1663/10819/16460
CONTEXT:  xlog redo Insert item, node: 1663/10819/16460 blkno: 514 offset: 78 
nitem: 1 isdata: F isleaf F isdelete F updateBlkno:1551

Now I'm working on fixing it.
-- 
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
                                                    WWW: http://www.sigaev.ru/

Attachments:

fix_ginstate_usage.patch (text/plain, 2.8K) download *** ./src/backend/access/gin.orig/ginentrypage.c Thu Oct 25 22:11:56 2007 --- ./src/backend/access/gin/ginentrypage.c Thu Oct 25 22:43:17 2007 *************** *** 404,409 **** --- 404,434 ---- } /* + * Returns new tuple with copied value from source tuple. + * New tuple will not store posting list + */ + static IndexTuple + copyIndexTuple(IndexTuple itup, Page page) + { + IndexTuple nitup; + + if (GinPageIsLeaf(page) && !GinIsPostingTree(itup)) + { + nitup = (IndexTuple) palloc(MAXALIGN(GinGetOrigSizePosting(itup))); + memcpy(nitup, itup, GinGetOrigSizePosting(itup)); + nitup->t_info &= ~INDEX_SIZE_MASK; + nitup->t_info |= GinGetOrigSizePosting(itup); + } + else + { + nitup = (IndexTuple) palloc(MAXALIGN(IndexTupleSize(itup))); + memcpy(nitup, itup, IndexTupleSize(itup)); + } + + return nitup; + } + + /* * Place tuple and split page, original buffer(lbuf) leaves untouched, * returns shadow page of lbuf filled new data. * Tuples are distributed between pages by equal size on its, not *************** *** 424,431 **** IndexTuple itup, leftrightmost = NULL; static ginxlogSplit data; - Datum value; - bool isnull; Page page; Page lpage = GinPageGetCopyPage(BufferGetPage(lbuf)); Page rpage = BufferGetPage(rbuf); --- 449,454 ---- *************** *** 494,502 **** ptr += MAXALIGN(IndexTupleSize(itup)); } ! value = index_getattr(leftrightmost, FirstOffsetNumber, btree->ginstate->tupdesc, &isnull); ! btree->entry = GinFormTuple(btree->ginstate, value, NULL, 0); ItemPointerSet(&(btree->entry)->t_tid, BufferGetBlockNumber(lbuf), InvalidOffsetNumber); btree->rightblkno = BufferGetBlockNumber(rbuf); data.node = btree->index->rd_node; --- 517,525 ---- ptr += MAXALIGN(IndexTupleSize(itup)); } ! btree->entry = copyIndexTuple(leftrightmost, lpage); ItemPointerSet(&(btree->entry)->t_tid, BufferGetBlockNumber(lbuf), InvalidOffsetNumber); + btree->rightblkno = BufferGetBlockNumber(rbuf); data.node = btree->index->rd_node; *************** *** 533,552 **** Page page = BufferGetPage(buf); itup = getRightMostTuple(page); ! if (GinPageIsLeaf(page) && !GinIsPostingTree(itup)) ! { ! nitup = (IndexTuple) palloc(MAXALIGN(GinGetOrigSizePosting(itup))); ! memcpy(nitup, itup, GinGetOrigSizePosting(itup)); ! nitup->t_info &= ~INDEX_SIZE_MASK; ! nitup->t_info |= GinGetOrigSizePosting(itup); ! } ! else ! { ! nitup = (IndexTuple) palloc(MAXALIGN(IndexTupleSize(itup))); ! memcpy(nitup, itup, IndexTupleSize(itup)); ! } ! ItemPointerSet(&nitup->t_tid, BufferGetBlockNumber(buf), InvalidOffsetNumber); return nitup; } --- 556,564 ---- Page page = BufferGetPage(buf); itup = getRightMostTuple(page); ! nitup = copyIndexTuple(itup, page); ItemPointerSet(&nitup->t_tid, BufferGetBlockNumber(buf), InvalidOffsetNumber); + return nitup; }

view thread (14+ messages)  latest in thread

Message-ID: <4721E335.8090601@sigaev.ru>
Permalink:  ../4721E335.8090601@sigaev.ru/
Also on:    postgresql.org/message-id/4721E335.8090601@sigaev.ru

 ·  · 

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: pgsql-admin@postgresql.org
  Cc: teodor@sigaev.ru, craig.mcelroy@contegix.com, tgl@sss.pgh.pa.us
  Subject: Re: Problem with PITR Past Particular WAL File
  In-Reply-To: <4721E335.8090601@sigaev.ru>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox