Comments on: How Patroni Addresses the Problem of the Logical Replication Slot Failover in a PostgreSQL Cluster

By: Shaan

Shaan — Mon, 15 Apr 2024 15:32:07 +0000

In reply to Jobin Augustine. below comment can be ignored and removed.

By: Shaan

Shaan — Fri, 12 Apr 2024 16:43:36 +0000

By: Stefan

Stefan — Thu, 14 Dec 2023 12:27:35 +0000

I’m stuck on an issue with Patroni and logical replication. I’ve been able to set all this up but now when logical replication starts I get a “publication does not exist” error. The publication does exist, in fact! The same case is documented here by someone else: https://postgrespro.com/list/thread-id/2492943 But there is no solution on the post. From what I understand, if you create the slot before your create the publication, then the slot does not know about the existence of the publication. I do not have control on when Patroni creates the slot and everything I have tried so far creates the publication after the slot is created, because I deploy the table/publication creation scripts after starting a new Patroni cluster on Kubernetes. Does anyone know how to resolve this?

By: Dominik Kroutvar

Dominik Kroutvar — Thu, 16 Nov 2023 09:37:01 +0000

Great post! I have set up a new Postgres cluster on my openshift cluster with 1 master and 2 replicas according to this recommendation. I have also configured 1 replication slot. Now I would like to monitor the health of the cluster. Therefore I have created an alert to fire if pg_exporter_last_scrape_error returns a value greater than 0. This metric works on the master as expected but all replicas return 1 instead of 0. Is this intended? Am I making a mistake in my thinking? Are there any tips on how and what is best to monitor in this scenario?

By: Bert

Bert — Thu, 24 Nov 2022 17:03:47 +0000

It's a nice feature! Using this feature requires adding slot information to the patroni.yml configuration file in advance. If the user add a subscription(include slot) without pre-creating replication slot in this way,What's a good way for me to find it in time and add it to the configuration file to avoid the loss of replication slot information after switching.Thanks!

By: Jobin Augustine

Jobin Augustine — Sat, 05 Feb 2022 07:08:23 +0000

In reply to Deepak. Hi Deepak, In this post, I am discussing about built-in native logical replication in PostgreSQL and slot maintenance for replication to downstream systems. No need to create any extension for it.

By: Deepak

Deepak — Thu, 03 Feb 2022 12:33:27 +0000

I had setup like this but my logical replication is not working .Reason being ,Patorni’s physical slot already in place .

Please note I have already set all parameters like wal_levle= logical and shared_preload_lib as well to logical and created extension as well but still logical replication is not working . Whenever i’m trying to create table on STANDBY ,its saying can’t create d/table in read only transaction

By: Jobin Augustine

Jobin Augustine — Wed, 26 Jan 2022 11:49:59 +0000

In reply to pcpg. Conceptually Yes, Standby can serve the request. But If you are thinking about leveraging standby as a publisher, it won't be possible now because logical decoding is not yet allowed on standby.

By: pcpg

pcpg — Tue, 25 Jan 2022 15:48:54 +0000

In reply to Jobin Augustine. Sorry, I realised that my query was not what I actually wanted to ask. Let me re-ask: so the standby needs to be a synchronous (streaming replication) standby. A change on primary, that is acknowledged to the client (application), will definitely be available on the standby and because the slot is maintained on the standby too, the subscriber can request the standby from the point till which replication was done. Correct?

By: Jobin Augustine

Jobin Augustine — Tue, 25 Jan 2022 15:25:54 +0000

In reply to pcpg. You are welcome :) . No need to worry even if slot on the standby lags. because it is the subscriber who makes the request so the slot will be automatically advanced.

By: Jobin Augustine

Jobin Augustine — Tue, 25 Jan 2022 14:57:56 +0000

In reply to Michael Banck. Sorry @Michael for being late. Yes Thats indeed a great idea. That can eliminate use of HAProxy Being a libpq connection that should work :) . But I am yet to test

By: Michael Banck

Michael Banck — Tue, 18 Jan 2022 11:02:26 +0000

Great post! Do you know whether it is possible to write all the hostnames into the subscriber connection string and target_session_attrs=read-write, so that the subscription would be reset to the new primary automatically?

By: pcpg

pcpg — Sat, 15 Jan 2022 04:45:07 +0000

Good to know this development by Patroni. Thanks for sharing in such detail. Could there be a scenario where the primary goes down before the slot is advanced on standby?