Horton

Re: ALTER TABLE .. DETACH PARTITION CONCURRENTLY

Alvaro Herrera <alvherre@alvh.no-ip.org>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Amit Langote <amitlangote09@gmail.com>
Cc: Justin Pryzby <pryzby@telsasoft.com>, Tom Lane <tgl@sss.pgh.pa.us>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2021-04-20T22:41:07Z
Lists: pgsql-hackers

Attachments

OK so after mulling this over for a long time, here's a patch for this.
It solves the problem by making the partition descriptor no longer be
cached if a detached partition is omitted.  Any transaction that needs a
partition descriptor that excludes detached partitions, will have to
recreate the partdesc from scratch.  To support this, I changed the
output boolean semantics: instead of "does this partdesc include an
detached partitions" as in your patch, it now is "are there any detached
partitions".  But whenever no detached partitions exist, or when all
partitions including detached are requested, then the cached descriptor
is returned (because that necessarily has to be correct).  The main
difference this has to your patch is that we always keep the descriptor
in the cache and don't rebuild anything, unless a detached partition is
present.

I flipped the find_inheritance_children() input boolean, from
"include_detached" to "omit_detached".  This is more natural, given the
internal behavior.  You could argue to propagate that naming change to
the partdesc.h API and PartitionDirectory, but I don't think there's a
need for that.

I ran all the detach-partition-concurrently tests under
debug_invalidate_system_caches_always=1 and everything passes.

I experimented with keeping a separate cached partition descriptor that
omits the detached partition, but that brings back some trouble; I
couldn't find a way to invalidate such a cached entry in a reasonable
way.  I have the patch for that, if somebody wants to play with it.

-- 
Álvaro Herrera       Valdivia, Chile
"That sort of implies that there are Emacs keystrokes which aren't obscure.
I've been using it daily for 2 years now and have yet to discover any key
sequence which makes any sense."                        (Paul Thomas)

Commits

  1. Make detach-partition-concurrently-4 less timing sensitive

  2. Improve documentation on DETACH PARTITION lock levels

  3. Track detached partitions more accurately in partdescs

  4. Allow a partdesc-omitting-partitions to be cached

  5. Fix relcache inconsistency hazard in partition detach

  6. Don't add a redundant constraint when detaching a partition

  7. ALTER TABLE ... DETACH PARTITION ... CONCURRENTLY

  8. Let ALTER TABLE Phase 2 routines manage the relation pointer

  9. Check default partitions constraints while descending