Re: [HACKERS] Runtime Partition Pruning

On 20 February 2018 at 23:46, Rajkumar Raghuwanshi
<rajkumar(dot)raghuwanshi(at)enterprisedb(dot)com> wrote:
> I have applied v10 patch on Amit's v27 over head ad7dbee36. I got "ERROR:
> partition missing from Append subplans" with the patch. on head and only
> with Amit's patches query is working fine, Please find test case below.

Thanks for the test case. I can recreate locally. This is down to the
fact that make_partition_pruneinfo() only makes sub-PartitionPruneInfo
for actual subpaths found in the Append. Your test case happens to
match both the part_p1 and part_p2 partitions on the first level of
iteration, but since no child of part_p2 is found in
make_partition_pruneinfo, that element in the subpartindex never gets
set.

The fix might be to just remove the error and silently ignore those
cases, but I was hoping to keep that around as it might catch other
bugs. I'm just not sure yet how to do both.

I'll rebase this on Amit's latest patch and have a think about it
while doing that.

--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services