Re: [HACKERS] Determine state of cluster (HA) - Mailing list pgsql-hackers

From Magnus Hagander
Subject Re: [HACKERS] Determine state of cluster (HA)
Date
Msg-id CABUevEz_fRkAm71+AarUYA+Kj6iV+yZSG2tLhjqQnwDYwkN9_Q@mail.gmail.com
Whole thread Raw
In response to Re: [HACKERS] Determine state of cluster (HA)  (Craig Ringer <craig@2ndquadrant.com>)
Responses Re: [HACKERS] Determine state of cluster (HA)
List pgsql-hackers


On Mon, Oct 16, 2017 at 4:39 AM, Craig Ringer <craig@2ndquadrant.com> wrote:
On 13 October 2017 at 08:50, Joshua D. Drake <jd@commandprompt.com> wrote:
> 5.  There is no way to connect to a db node with something akin to
> SQL-Server's "application intent" flags, to allow a connection to be
> rejected if we wish it to be a read/write connection.  This helps detect the
> state of the node directly without having to ask any further questions of
> the node, and makes it easier to "stall" during connection until a proper
> connection can be made.

That sounds desirable, and a good step toward eventually being able to
transparently re-route read/write queries from replica to master.
Which is where I'd like to land up eventually.

It also sounds a lot like the connection parameter target_session_attrs, does it not? We don't reroute active connections based on it, and we're not smart enough to do anything beyond "try them one by one until you reach the one with the correct attributes", but the basic functionality is there. Basically what we already have fulfills what JD is suggesting, but not what Craig is, if I understand it correctly.

--

pgsql-hackers by date:

Previous
From: Michael Paquier
Date:
Subject: Re: [HACKERS] relkind check in DefineIndex
Next
From: Amit Kapila
Date:
Subject: Re: [HACKERS] Parallel safety for extern params