Re: Libpq support to connect to standby server as priority

From: Michael Paquier <michael(at)paquier(dot)xyz>
To: "Tsunakawa, Takayuki" <tsunakawa(dot)takay(at)jp(dot)fujitsu(dot)com>
Cc: 'Tom Lane' <tgl(at)sss(dot)pgh(dot)pa(dot)us>, Dave Cramer <pg(at)fastcrypt(dot)com>, Robert Haas <robertmhaas(at)gmail(dot)com>, "laurenz(dot)albe(at)cybertec(dot)at" <laurenz(dot)albe(at)cybertec(dot)at>, Haribabu Kommi <kommi(dot)haribabu(at)gmail(dot)com>, Jing Wang <jingwangian(at)gmail(dot)com>, PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Libpq support to connect to standby server as priority
Date: 2019-01-15 02:48:55
Message-ID: 20190115024855.GC1433@paquier.xyz
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Tue, Jan 15, 2019 at 02:00:57AM +0000, Tsunakawa, Takayuki wrote:
> But as some people said, I don't think this is the right way. I
> suppose what's leading to the current somewhat complicated situation
> is that there was no easy way for the client to know whether the
> server is the master. That ended up in using "SHOW
> transaction_read_only" instead, and people supported that compromise
> by saying "read only status is more useful than whether the server
> is standby or not," I'm afraid.

Right. Another pin point is that it is complicated for a client to be
sure that it is connected to a standby as at the time between
transaction_read_only is checked and the connection is reported as
ready to be used for the application, you may be actually linked to a
primary which has just recently been promoted. I am not personally
sure if it is worth caring about that in such level of details to get
to get something useful, but there have been doubts about not making
that absolutely right to leverage correctly applications willing to
use read-only clients.

> The original desire should have been the ability to connect to a
> primary or a standby. So, I think we should go back to the original
> thinking (and not complicate the feature), and create a read only
> GUC_REPORT variable, say, server_role, that identifies whether the
> server is a primary or a standby.

From the point of view of making sure that a client is really
connected to a primary or a standby, this is the best idea around.
--
Michael

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tsunakawa, Takayuki 2019-01-15 02:52:38 RE: Protect syscache from bloating with negative cache entries
Previous Message Andres Freund 2019-01-15 02:46:49 Re: strange valgrind failures (again)