Horton

Re: BUG #17949: Adding an index introduces serialisation anomalies.

Dmitry Dolgov <9erthalion6@gmail.com>

View thread View on postgr.es
From: Dmitry Dolgov <9erthalion6@gmail.com>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: artem.anisimov.255@gmail.com, pgsql-bugs@lists.postgresql.org, Teodor Sigaev <teodor@sigaev.ru>, Heikki Linnakangas <hlinnaka@iki.fi>
Date: 2023-06-21T09:04:21Z
Lists: pgsql-bugs 
> On Wed, Jun 21, 2023 at 11:20:28AM +1200, Thomas Munro wrote:
> On Wed, Jun 21, 2023 at 10:54 AM Thomas Munro <thomas.munro@gmail.com> wrote:
> > This is probably going to go faster if I CC the authors of commit
> > 0bef1c06.  Any ideas about how we're missing rw-conflicts under high
> > concurrency?
>
> I guess one (probably stupid) question I have: where is the
> CheckForSerializableConflictIn(rootPostingTree) that I was expecting
> to see when writing to an existing posting tree?  IIUC that should
> pair with PredicateLockPage(btree->index, rootPostingTree, snapshot)
> when reading.

I can't find it either, but based on my superficial investigation this
particular reproducer doesn't seem to hit posting trees functionality at
all. What I observe is the inserting transaction uses
ginHeapTupleFastCollect + ginHeapTupleFastInsert, and the corresponding
commentary says that serialization in this case depends on the
metabuffer:

	/*
	 * An insertion to the pending list could logically belong anywhere in the
	 * tree, so it conflicts with all serializable scans.  All scans acquire a
	 * predicate lock on the metabuffer to represent that.
	 */

Now the reading transaction actually does PredicateLockPage on the
metabuffer inside scanPendingInsert, but strangely enough it doesn't
lock anything because the SerializationNeededForRead condition is false.
I'm trying to verify if it's somehow a part of the issue, or something
is broken on my side.

Commits

  1. Fix race in SSI interaction with bitmap heap scan.

  2. Fix race in SSI interaction with gin fast path.

  3. Fix race in SSI interaction with empty btrees.

  4. Re-think predicate locking on GIN indexes.