Re: Does a row lock taken out in a CTE stay in place?

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Seamus Abshere <seamus(at)abshere(dot)net>
Cc: pgsql-general(at)postgresql(dot)org
Subject: Re: Does a row lock taken out in a CTE stay in place?
Date: 2017-07-11 15:45:45
Message-ID: 1604.1499787945@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

Seamus Abshere <seamus(at)abshere(dot)net> writes:
> Given an update that uses CTEs like this:
> WITH
> lock_rows AS (
> SELECT 1 FROM tbl WHERE [...] FOR UPDATE
> )
> UPDATE [...]

> Will the rows in `tbl` remain locked until the UPDATE is finished?

Yes, locks are associated with a transaction not a statement or
sub-statement.

> Also, does it matter if `lock_rows` is referenced? (IIUC the query
> wouldn't be run if the CTE isn't referenced if it was for a SELECT, but
> since it's an UPDATE, it will be run anyway)

Yes, it does --- unreferenced SELECT CTEs are discarded. I thought maybe
there was an exception for FOR UPDATE, but a look at the code says
differently. In any case we would only lock rows the sub-select had
actually read, so if it's not called by the outer statement it would
still be a no-op.

regards, tom lane

In response to

Browse pgsql-general by date

  From Date Subject
Next Message David G. Johnston 2017-07-11 15:47:09 Re: Does a row lock taken out in a CTE stay in place?
Previous Message Seamus Abshere 2017-07-11 15:36:17 Does a row lock taken out in a CTE stay in place?