Extend more usecase for planning time partition pruning and init partition pruning. - Mailing list pgsql-hackers

From Andy Fan
Subject Extend more usecase for planning time partition pruning and init partition pruning.
Date
Msg-id CAKU4AWq4NLxu5JF9+d=o=A636-=eFNFmPx+kJ44ezTm=ikZ73w@mail.gmail.com
Whole thread Raw
Responses Re: Extend more usecase for planning time partition pruning and init partition pruning.  (Andy Fan <zhihui.fan1213@gmail.com>)
List pgsql-hackers
Hi:

 I recently found a use case like this.  SELECT * FROM p, q WHERE p.partkey =
 q.colx AND (q.colx = $1 OR q.colx = $2); Then we can't do either planning time
 partition prune or init partition prune.  Even though we have run-time
 partition pruning work at last, it is too late in some cases since we have
 to init all the plan nodes in advance.  In my case, there are 10+
 partitioned relation in one query and the execution time is short, so the
 init plan a lot of plan nodes cares a lot.

The attached patches fix this issue. It just get the "p.partkey = q.colx"
case in root->eq_classes or rel->joinlist (outer join), and then check if there
is some baserestrictinfo in another relation which can be used for partition
pruning. To make the things easier, both partkey and colx must be Var
expression in implementation.

- v1-0001-Make-some-static-functions-as-extern-and-extend-C.patch

Just some existing refactoring and extending ChangeVarNodes to be able
to change var->attno.

- v1-0002-Build-some-implied-pruning-quals-to-extend-the-us.patch

Do the real job.

Thought?



--
Best Regards
Attachment

pgsql-hackers by date:

Previous
From: Andy Fan
Date:
Subject: Re: [PATCH] Keeps tracking the uniqueness with UniqueKey
Next
From: "Joel Jacobson"
Date:
Subject: Re: [HACKERS] GSoC 2017: Foreign Key Arrays