Date: Fri, 4 Dec 2020 11:37:20 -0300
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Dmitry Dolgov <9erthalion6@gmail.com>
Cc: Pg Hackers <pgsql-hackers@lists.postgresql.org>,
	Michael Paquier <michael@paquier.xyz>,
	James Coleman <jtc331@gmail.com>
Subject: Re: {CREATE INDEX, REINDEX} CONCURRENTLY improvements
Message-ID: <20201204143720.GA32678@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201204113832.yuneqs6t4fhjo6k3@localhost>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On 2020-Dec-04, Dmitry Dolgov wrote:

> * This one is mostly for me to understand. There are couple of places
>   with a commentary that 'PROC_IN_SAFE_IC is not necessary, because the
>   transaction only takes a snapshot to do some catalog manipulation'.
>   But for some of them I don't immediately see in the relevant code
>   anything related to snapshots. E.g. one in DefineIndex is followed by
>   WaitForOlderSnapshots (which seems to only do waiting, not taking a
>   snapshot), index_set_state_flags and CacheInvalidateRelcacheByRelid.
>   Is taking a snapshot hidden somewhere there inside?

Well, they're actually going to acquire an Xid, not a snapshot, so the
comment is slightly incorrect; I'll fix it, thanks for pointing that
out.  The point stands: because those transactions are of very short
duration (at least of very short duration after the point where the XID
is obtained), it's not necessary to set the PROC_IN_SAFE_IC flag, since
it won't cause any disruption to other processes.

It is possible that I copied the wrong comment in DefineIndex.  (I only
noticed that one after I had mulled over the ones in
ReindexRelationConcurrently, each of which is skipping setting the flag
for slightly different reasons.)