From: Tom Lane <tgl@sss.pgh.pa.us>
To: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
cc: Andres Freund <andres@anarazel.de>,
        PostgreSQL-development <pgsql-hackers@postgresql.org>,
        Alvaro Herrera <alvherre@2ndquadrant.com>,
        Bruce Momjian <bruce@momjian.us>,
        Michael Paquier <michael@paquier.xyz>
Subject: Re: transction_timestamp() inside of procedures
In-reply-to: <38e36919-8bee-c717-be99-91137ad28aaf@2ndquadrant.com>
References: <20180926155443.u4zh7jactzflys3z@alvherre.pgsql>
 <ec816bd0-0357-23a9-bea2-3ef1619eb340@2ndquadrant.com>
 <e2ad8aaf-e1d8-499e-4de0-e3616f8d908d@2ndquadrant.com>
 <20181002145853.wvhcb7pbrzgjyxns@alap3.anarazel.de>
 <38e36919-8bee-c717-be99-91137ad28aaf@2ndquadrant.com>
Comments: In-reply-to Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
	message dated "Fri, 05 Oct 2018 13:28:54 +0200"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <27413.1539018874.1@sss.pgh.pa.us>
Date: Mon, 08 Oct 2018 13:14:34 -0400
Message-ID: <27414.1539018874@sss.pgh.pa.us>
Precedence: bulk

Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:
> On 02/10/2018 16:58, Andres Freund wrote:
>> It's a bit weird to make this decision based on these two timestamps
>> differing.  For one, it only indirectly seems to be guaranteed that
>> xactStartTimestamp is even set to anything here (to 0 by virtue of being
>> a global var).

> Maybe but it seems to be the simplest way without doing major surgery to
> all this code.

This patch doesn't apply over 07ee62ce9.  Also, I like the
timestamp-comparison approach even less than Andres does: I think it's
probably outright broken, especially since it treats the equality case
as license to advance xactStartTimestamp.

Surely there is some way that we can directly test whether we're inside a
procedure or not?  I think the logic should be basically

    if (!IsParallelWorker())
+   {
+       if (!InsideProcedure())
            xactStartTimestamp = stmtStartTimestamp;
+       else
+           xactStartTimestamp = GetCurrentTimestamp();
+   }
    else
       Assert(xactStartTimestamp != 0);


			regards, tom lane