User-Agent: Cyrus-JMAP/3.5.0-alpha0-273-g8500d2492d-fm-20210323.002-g8500d249
Mime-Version: 1.0
Message-Id: <42a0fd97-8cb2-4615-b9a8-698dc7c55962@www.fastmail.com>
In-Reply-To: <3969078.1618532221@sss.pgh.pa.us>
References: <eae3cfcd-fe79-4516-b0ff-3872f2880943@www.fastmail.com>
 <3969078.1618532221@sss.pgh.pa.us>
Date: Fri, 16 Apr 2021 16:16:45 +1000
From: "Daniel Heath" <daniel@heath.cc>
To: "Tom Lane" <tgl@sss.pgh.pa.us>
Cc: pgsql-novice@lists.postgresql.org
Subject: Re: Prepared statement invalidation
Content-Type: text/plain
Archived-At: 
 <https://www.postgresql.org/message-id/42a0fd97-8cb2-4615-b9a8-698dc7c55962%40www.fastmail.com>
Precedence: bulk

Is there a transaction isolation mode that would help with the specific case of adding a column to a table? EG to prevent a prepared statement from seeing the new schema?

Thanks,
Daniel Heath

On Fri, Apr 16, 2021, at 10:17 AM, Tom Lane wrote:
> "Daniel Heath" <daniel@heath.cc> writes:
> > I've found that when I add a new column, I sometimes get a spike of errors on the application side due to prepared statement query plans getting invalidated, causing transactions to rollback.
> > The error is:
> > ERROR: cached plan must not change result type 
> > Is there a way to make postgres recalculate the query plan instead of failing the transaction when a new column is added?
> 
> No.  It'd be easy enough to remove that restriction on the server side,
> but we're afraid that it would break applications, which probably aren't
> expecting the result rowtype of a prepared query to be different from what
> they were told (perhaps only milliseconds earlier).
> 
> I'd say the short answer is "don't use prepared queries, or if you do,
> spell out the columns you want instead of saying SELECT *".
> 
> 			regards, tom lane
> 
> 
> 
> 
> > Thanks,
> > Daniel Heath
> 
>