Re: [HACKERS] path toward faster partition pruning

Amit Langote <langote_amit_f8@lab.ntt.co.jp>

From: Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>

To: David Rowley <david.rowley@2ndquadrant.com>

Cc: Robert Haas <robertmhaas@gmail.com>, Ashutosh Bapat <ashutosh.bapat@enterprisedb.com>, Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>, Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp>, Jesper Pedersen <jesper.pedersen@redhat.com>, Amit Langote <amitlangote09@gmail.com>, Dilip Kumar <dilipbalaut@gmail.com>, Beena Emerson <memissemerson@gmail.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>

Date: 2018-03-26T11:42:57Z

Lists: pgsql-hackers

Attachments

v40-0001-Add-partsupfunc-to-PartitionSchemeData.patch (text/plain) patch v40-0001
v40-0002-Add-more-tests-for-partition-pruning.patch (text/plain) patch v40-0002
v40-0003-Faster-partition-pruning.patch (text/plain) patch v40-0003
v40-0004-Add-only-unpruned-partitioned-child-rels-to-part.patch (text/plain) patch v40-0004

Hi David.

On 2018/03/25 14:32, David Rowley wrote:
> On 25 March 2018 at 18:28, David Rowley <david.rowley@2ndquadrant.com> wrote:
>> The attached delta applies on top of v39 plus delta1.
> 
> Sorry, the attached should do this. Ignore the last attachment.

I have incorporated both of your delta1 and delta2_1 patches.

Your proposed change to determine the cross-type comparison function OID
during planning itself is a good one, although I wasn't sure why it was
done only for the <> operators.  I also implemented that for
PartitionPruneStepOp steps.

Also, I started thinking that implementing pruning using <> operators with
a PartitionPruneCombineOp was not such a great idea.  That needed us to
add argexprs and argcmpfns to that struct, which seemed a bit odd.  I
defined a new pruning node type called PartitionPruneStepOpNe, which still
seems a bit odd, but given that our support for pruning using <> is quite
specialized, that may be fine.

I added a bunch of hopefully informative comments in partprune.c and for
the struct definitions of pruning step nodes.

Please find attached find a new version.

Thanks,
Amit

Commits

Fix assorted partition pruning bugs
- d758d9702e2f 11.0 landed
Make gen_partprune_steps static
- d1e2cac5ff7e 11.0 landed
Remove useless 'default' clause
- c775fb9e18ac 11.0 landed
Reorganize partitioning code
- da6f3e45ddb6 11.0 landed
Use custom hash opclass for hash partition pruning
- fafec4cce814 11.0 landed
Blindly attempt to fix sepgsql tests broken due to 9fdb675fc5.
- 4f813c7203e0 11.0 landed
Attempt to fix endianess issues in new hash partition test.
- 40e42e1024c5 11.0 landed
Faster partition pruning
- 9fdb675fc5d2 11.0 landed
For partitionwise join, match on partcollation, not parttypcoll.
- 2af28e603319 11.0 landed
Revise API for partition bound search functions.
- f724022d0ae0 11.0 landed
Revise API for partition_rbound_cmp/partition_rbound_datum_cmp.
- b0229235564f 11.0 landed
Fix possible crash in partition-wise join.
- f069c91a5793 11.0 cited
Refactor code for partition bound searching
- 9aef173163ae 11.0 landed
New C function: bms_add_range
- 84940644de93 11.0 landed
Add extensive tests for partition pruning.
- 8d4e70a63bf8 11.0 landed
Add null test to partition constraint for default range partitions.
- 7b88d63a9122 11.0 cited
Remove BufFile's isTemp flag.
- 11e264517dff 11.0 cited
Make OWNER TO subcommand mention consistent
- bf54c0f05c0a 11.0 cited
Fix index matching for operators with mixed collatable/noncollatable inputs.
- cb37c291060d 9.2.0 cited