Message-ID: <8b788512e88c21ce5e913963827de3e3710d245e.camel@cybertec.at>
Subject: Re: Table-Level Locks and Transactions
From: Laurenz Albe <laurenz.albe@cybertec.at>
To: Simon Riggs <simon.riggs@enterprisedb.com>, greg.rychlewski@gmail.com
Cc: pgsql-novice@lists.postgresql.org
Date: Thu, 09 Dec 2021 12:47:13 +0100
In-Reply-To: 
 <CANbhV-EyDBV_z5urPQHshMakkScUwRdfBjektEB3NYkdmc4=1g@mail.gmail.com>
References: 
	<CAKemG7XD-UU7AGan+jp5jtw+6wwcN4z3pPSB=DuoAC1z3JDDhg@mail.gmail.com>
	 <CANbhV-EyDBV_z5urPQHshMakkScUwRdfBjektEB3NYkdmc4=1g@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.40.4 (3.40.4-2.fc34) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Archived-At: 
 <https://www.postgresql.org/message-id/8b788512e88c21ce5e913963827de3e3710d245e.camel%40cybertec.at>
Precedence: bulk

On Thu, 2021-12-09 at 10:33 +0000, Simon Riggs wrote:
> On Thu, 9 Dec 2021 at 01:07, Greg Rychlewski <greg.rychlewski@gmail.com> wrote:
> > 
> > I'm curious to understand when table-level locks are released by one transaction to use in another.
> 
> At the end of transaction.
> 
> > For instance, say I have the following transaction, where index1 and index2 both belong to the same table.
> > 
> > BEGIN;
> > DROP INDEX index1;
> > CREATE INDEX index2;
> > COMMIT;
> > 
> > This transaction acquires an ACCESS EXCLUSIVE lock during the DROP INDEX step and then acquires a SHARE lock during the CREATE INDEX step.
> > 
> > I understand that there will be no conflicts between the statements in this transaction,
> > but I'm wondering if an outside transaction will think that the table has an ACCESS EXCLUSIVE
> > lock until this entire transaction is completed.
> 
> No, just index1, but the SHARE lock will prevent writes.

Not sure what you mean by that, but the ACCESS EXCLUSIVE lock from DROP INDEX
will be held until the end of the transaction.


> Have you thought about using DROP INDEX CONCURRENTLY and CREATE INDEX
> CONCURRENTLY?

These cannot be used in a transaction, however.

> If in doubt, try it. Look at pg_locks.

+1

Yours,
Laurenz Albe
-- 
Cybertec | https://www.cybertec-postgresql.com