Re: Declarative partitioning - another take

Hi Andres,

On 2017/01/20 15:15, Andres Freund wrote:
> On 2017-01-19 14:18:23 -0500, Robert Haas wrote:
>> Committed.
>
> One of the patches around this topic committed recently seems to cause
> valgrind failures like
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=skink&dt=2017-01-19%2008%3A40%3A02
> :
> ==24969== Conditional jump or move depends on uninitialised value(s)
> ==24969== at 0x4C38DA: btint4cmp (nbtcompare.c:97)
> ==24969== by 0x83860C: FunctionCall2Coll (fmgr.c:1318)
> ==24969== by 0x536643: partition_bounds_equal (partition.c:627)
> ==24969== by 0x820864: equalPartitionDescs (relcache.c:1203)
> ==24969== by 0x82423A: RelationClearRelation (relcache.c:2553)
> ==24969== by 0x8248BA: RelationFlushRelation (relcache.c:2662)
> ==24969== by 0x824983: RelationCacheInvalidateEntry (relcache.c:2714)
> ==24969== by 0x81D9D6: LocalExecuteInvalidationMessage (inval.c:568)
> ==24969== by 0x81CB0D: ProcessInvalidationMessages (inval.c:444)
> ==24969== by 0x81D3CB: CommandEndInvalidationMessages (inval.c:1056)
> ==24969== by 0x4F6735: AtCCI_LocalCache (xact.c:1374)
> ==24969== by 0x4F8249: CommandCounterIncrement (xact.c:957)
> ==24969== Uninitialised value was created by a heap allocation
> ==24969== at 0x85AA83: palloc (mcxt.c:914)
> ==24969== by 0x53648E: RelationBuildPartitionDesc (partition.c:528)
> ==24969== by 0x823F93: RelationBuildDesc (relcache.c:1348)
> ==24969== by 0x8241DB: RelationClearRelation (relcache.c:2524)
> ==24969== by 0x8248BA: RelationFlushRelation (relcache.c:2662)
> ==24969== by 0x824983: RelationCacheInvalidateEntry (relcache.c:2714)
> ==24969== by 0x81D9D6: LocalExecuteInvalidationMessage (inval.c:568)
> ==24969== by 0x81CB0D: ProcessInvalidationMessages (inval.c:444)
> ==24969== by 0x81D3CB: CommandEndInvalidationMessages (inval.c:1056)
> ==24969== by 0x4F6735: AtCCI_LocalCache (xact.c:1374)
> ==24969== by 0x4F8249: CommandCounterIncrement (xact.c:957)
> ==24969== by 0x82538B: RelationSetNewRelfilenode (relcache.c:3490)
> ==24969==
> ==24969== VALGRINDERROR-END

Thanks for the report. This being my first time reading a valgrind report
on buildfarm, is it correct to to assume that the command immediately
preceding VALGRINDERROR-BEGIN is what triggered the failure?

... LOG: statement: truncate list_parted;
==24969== VALGRINDERROR-BEGIN
==24969== Conditional jump or move depends on uninitialised value(s)
==24969== at 0x4C38DA: btint4cmp (nbtcompare.c:97)

So in this case: truncate list_parted?

Thanks,
Amit