Horton

Re: Partition-wise join for join between (declaratively) partitioned tables

Dilip Kumar <dilipbalaut@gmail.com>

View thread View on postgr.es
From: Dilip Kumar <dilipbalaut@gmail.com>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Rafia Sabih <rafia.sabih@enterprisedb.com>, Ashutosh Bapat <ashutosh.bapat@enterprisedb.com>, Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>, Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2017-07-25T16:09:36Z
Lists: pgsql-hackers 
On Tue, Jul 25, 2017 at 8:59 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> On Tue, Jul 25, 2017 at 1:31 AM, Rafia Sabih
> <rafia.sabih@enterprisedb.com> wrote:
>> - other queries show a good 20-30% improvement in performance. Performance
>> numbers are as follows,
>>
>> Query| un_part_head (seconds) | part_head (seconds) | part_patch (seconds) |
>> 3 | 76 |127 | 88 |
>> 4 |17 | 244 | 41 |
>> 5 | 52 | 123 | 84 |
>> 7 | 73 | 134 | 103 |
>> 10 | 67 | 111 | 89 |
>> 12 | 53 | 114 | 99 |
>> 18 | 447 | 709 | 551 |
>
> Hmm.  This certainly shows that benefit of the patch, although it's
> rather sad that we're still slower than if we hadn't partitioned the
> data in the first place.  Why does partitioning hurt performance so
> much?

I was analysing some of the plans (without partition and with
partition), Seems like one of the reasons of performing worse with the
partitioned table is that we can not use an index on the partitioned
table.

Q4 is taking 17s without partition whereas it's taking 244s with partition.

Now if we analyze the plan

Without partition, it can use parameterize index scan on lineitem
table which is really big in size. But with partition, it has to scan
this table completely.

                          ->  Nested Loop Semi Join
                                 ->  Parallel Bitmap Heap Scan on orders
                                       ->  Bitmap Index Scan on
idx_orders_orderdate  (cost=0.00..24378.88 r
                       ->  Index Scan using idx_lineitem_orderkey on
lineitem  (cost=0.57..29.29 rows=105 width=8) (actual
time=0.031..0.031 rows=1 loops=1122364)
                                       Index Cond: (l_orderkey =
orders.o_orderkey)
                                       Filter: (l_commitdate < l_receiptdate)
                                       Rows Removed by Filter: 1

-- 
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com

Commits

  1. Basic partition-wise join functionality.

  2. Assorted preparatory refactoring for partition-wise join.

  3. Teach adjust_appendrel_attrs(_multilevel) to do multiple translations.

  4. Stamp 10beta2.

  5. Eat XIDs more efficiently in recovery TAP test.

  6. Abstract logic to allow for multiple kinds of child rels.

  7. Implement SortSupport for macaddr data type

  8. Attempt to stabilize grouping sets regression test plans.

  9. Teach xlogreader to follow timeline switches

  10. Don't scan partitioned tables.

  11. Fix grammar.

  12. postgres_fdw: Push down FULL JOINs with restriction clauses.

  13. Some preliminary refactoring towards partitionwise join.

  14. contrib/amcheck needs RecentGlobalXmin to be PGDLLIMPORT'ified.

  15. Print test parameters like "foo: 123", and results like "foo = 123".