Horton

Re: [HACKERS] Runtime Partition Pruning

Robert Haas <robertmhaas@gmail.com>

View thread View on postgr.es
From: Robert Haas <robertmhaas@gmail.com>
To: David Rowley <david.rowley@2ndquadrant.com>
Cc: Beena Emerson <memissemerson@gmail.com>, Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>, amul sul <sulamul@gmail.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>, Dilip Kumar <dilipbalaut@gmail.com>
Date: 2017-12-22T01:29:46Z
Lists: pgsql-hackers 
On Thu, Dec 21, 2017 at 6:53 PM, David Rowley
<david.rowley@2ndquadrant.com> wrote:
> On 22 December 2017 at 12:45, Robert Haas <robertmhaas@gmail.com> wrote:
>> PREPARE q3 (int, int) AS SELECT * FROM unpartitioned_table WHERE
>> partkey BETWEEN $1 AND $2;
>>
>> ...that has to decide whether to use an index.  And to do that it has
>> to estimate what fraction of the table will match the BETWEEN clause.
>> That may be an uninformed guess, but it guesses something.  We
>> probably want to do something here that makes the guess for a
>> partitioned_table similar to the guess for an unpartitioned_table.
>
> Are you considering some sort of clauselist_selectivity() estimate on
> the given parameters and comparing that to the same selectivities that
> were determined for the previous custom plans?

No, I don't think comparing to previous custom plans is a workable
approach.  I was thinking, rather, that if we know for example that
we've doing pruning on partition_column = $1, then we know that only
one partition will match.  That's probably a common case.  If we've
got partition_column > $1, we could assume that, say, 75% of the
partitions would match.  partition_column BETWEEN $1 and $2 is
probably a bit more selective, so maybe we assume 50% of the
partitions would match.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Commits

  1. Initialize ExprStates once in run-time partition pruning

  2. Add bms_prev_member function

  3. Support partition pruning at execution time

  4. Document partprune.c a little better

  5. Fix crash in pg_replication_slot_advance

  6. Add parallel-aware hash joins.

  7. Fix typo.

  8. Allow --with-bonjour to work with non-macOS implementations of Bonjour.

  9. Mention need for --no-inc-recursive in rsync command