Dimitri Fontaine

“This book is on a higher level than generic software books. It explains the how's and why's of using Postgres's many feature. It is an excellent companion to the official Postgres documentation.” wrote Bruce Momjian about my book theartofpostgresql.com. You have your copy yet?

The state of the art Cardinality Estimation Algorithm called HyperLogLog is available as a @PostgreSQL extension! Read more about it at tapoueh.org/blog/2013/02/p…

Postgres 15 was released yesterday. Now that you've read the release notes and picked your favorite features, what questions can I answer for you?

How to quickly find data matching a complex set of user provided tags, all within @PostgreSQL query engine, you may ask! In the following article we use some denormalization tricks to answer that, read more at tapoueh.org/blog/2013/10/d…

@unamashana App devs are afraid of putting logic in SQL 🙂 @tapoueh ‘s The Art of PostgreSQL covers it well:

How to Write SQL as code and maintain it the same way as the rest of your code checkout? Read more at tapoueh.org/blog/2017/06/h…

@fuzzychef I suppose yes. Ideally we would have a quorum based approach for multiple standbies in that place of the code. Remember that disconnected nodes are not part of pg_stat_replication though, making our lives a little harder. Contributions welcomed!

I am proud to announce release 2.0 of pg_auto_failover! With Postgres 15 support, and already in debian thanks to @df7cb amazing work. Get started at pg-auto-failover.readthedocs.io/en/v2.0/index.… with the docs. You're looking to implement HA for Postgres? Try pg_auto_failover!

Postgres 15 is almost ready. Let's check some changes that might break compatibility with previous releases: 🧵

Final tweet in that thread: see #create-a-citus-cluster-take-two" target="_blank" rel="nofollow noopener">pg-auto-failover.readthedocs.io/en/v2.0/citus-… if you're wondering about “a single command line per box, really?”. Yes, really. And you can reproduce that at home. The full Dockerfile and docker-compose.yml setup is provided.

Implementing these reference architectures require a single command line per box in many cases. See the docker-compose files in the tutorials. More architectures are possible, too, with quite some flexibility!

@d_gustafsson What about always going through the dry-run checks first, with clear logging for diagnostic? And allowing the drop node command to finish the clean-up when required too. Maybe just fixing the drop node command is a good start…

@tapoueh I'm pondering a "create postgres --dry-run" to run through the motions and check for issues

@tapoueh any idea why I would get this in a clean VM with no ~/.config/ or datadir leftovers when creating a first node: ERROR NodeId, nodename, hostname or port returned by monitor is NULL ERROR Failed to set node 1 sysidentifier to "7152867231273596547" on the monitor becaus

@d_gustafsson We definitely need that for sure. I’m thinking of an additional feature in the same kind of area… we have some bug reports with a stale FSM. Rare, but happens. A way to re-Init would come in handy then. A total FSM prover also :-)

@tapoueh Or at least some more rigorous checking on "create postgres" to make sure nothing was left behind that can impact. I'll see what I can come up with

@d_gustafsson Yes please! Related: I am thinking we should have a ré-initié operation… maybe pg_auto to drop node —re-create or something. Would simplify things and also help unblock in some situations. What do you think?

@tapoueh turns out it was the ~/.local state files being left after a pg_autoctl drop node --destroy which then blocked it from creating a new node .. shouldn't that file be cleaned up as well? Should I hack up a patch to fix?