From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
cc: Michael Paquier <michael@paquier.xyz>, pgsql-hackers@lists.postgresql.org
Subject: Re: Gripes about walsender command processing
In-reply-to: <20200916221422.GA24095@alvherre.pgsql>
References: <20200916221422.GA24095@alvherre.pgsql>
Comments: In-reply-to Alvaro Herrera <alvherre@2ndquadrant.com>
	message dated "Wed, 16 Sep 2020 19:14:22 -0300"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <39193.1600294868.1@sss.pgh.pa.us>
Date: Wed, 16 Sep 2020 18:21:08 -0400
Message-ID: <39194.1600294868@sss.pgh.pa.us>
Precedence: bulk

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2020-Sep-16, Tom Lane wrote:
>> I don't see an easy way to improve on it though.  The only obvious
>> alternative would be to put another switch before the main one that
>> just fills a "const char *cmdtag" variable, but that seems ugly.

> The alternative of doing the assignment in each case of the same switch
> does not look too terrible:

>  		case T_IdentifySystemCmd:
> +			cmdtag = "IDENTIFY_SYSTEM";
> +			set_ps_display(cmdtag);
>  			IdentifySystem();
> +			EndReplicationCommand(cmdtag);
>  			break;
 
>  		case T_BaseBackupCmd:
> +			cmdtag = "BASE_BACKUP";
> +			set_ps_display(cmdtag);
>  			PreventInTransactionBlock(true, cmdtag);
>  			SendBaseBackup((BaseBackupCmd *) cmd_node);
> +			EndReplicationCommand(cmdtag);
>  			break;

Yeah, that works for me.  It doesn't allow for having just one
set_ps_display() call ahead of the switch, but that isn't that
big a loss.  We cannot merge the EndReplicationCommand calls to
after the switch, because some of the cases don't want one here;
so that partial duplication is inescapable.

Note that your changes need to be backpatched into v13,
because AFAICS this code is violating the FE/BE protocol
right now --- it's just luck that libpq isn't moaning
about extra CommandComplete messages.  But I don't think
we want to change the ps title behavior in v13 at this
late date, so that part should be HEAD-only.

			regards, tom lane