PartitionDispatch's partdesc field

From: Robert Haas <robertmhaas(at)gmail(dot)com>
To: "pgsql-hackers(at)postgresql(dot)org" <pgsql-hackers(at)postgresql(dot)org>
Subject: PartitionDispatch's partdesc field
Date: 2018-07-25 20:23:29
Message-ID: CA+TgmobFnxcaW-Co-XO8=yhJ5pJXoNkCj6Z7jm9Mwj9FGv-D7w@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

I noticed today that in the PartitionDispatch structure, the partdesc
field is set but not used. So we could remove it. See attached
pd-partdesc-remove.patch. If we want to go this route, I suggest
doing a slightly more thorough removal and getting rid of the key
field as well. See attached pd-partdesc-and-key-remove.patch.

Another alternative, which I think might make more sense, is to make
use pd->key and pd->partdesc in preference to pd->reldesc->rd_partkey
and pd->reldesc->rd_partdesc. It seems to me that the idea of the
PartitionDispatch structure is that it gathers together all of the
information that we need for tuple routing, so it might make sense for
the tuple routing code ought to get the information from there rather
than referring back to the RelationDesc. See attached
pd-partdesc-use.patch.

Basically, the decision here is whether we want to avoid invoking
RelationGetPartitionKey and RelationGetPartitionDesc over and over
again, either for reasons of notational cleanliness (macro names are
long) or safety (absolutely no chance that the answer will change) or
efficiency (maybe those macros will someday do more than just a
pointer dereference?). If so, we should cache the data in the
PartitionDispatch object and then use it. If not, there seems to be
no reason to copy those pointers from the Relation to the
PartitionDispatch in the first place, since the latter has a pointer
to the former anyway.

Opinions?

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

Attachment Content-Type Size
pd-partdesc-remove.patch application/octet-stream 1.2 KB
pd-partdesc-and-key-remove.patch application/octet-stream 3.1 KB
pd-partdesc-use.patch application/octet-stream 2.1 KB

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Robert Haas 2018-07-25 20:27:41 Re: Expression errors with "FOR UPDATE" and postgres_fdw with partition wise join enabled.
Previous Message Andres Freund 2018-07-25 20:15:33 Re: JIT breaks PostGIS