“Multi-AZ” in Amazon RDS and how it may differ from High Availability: https://dev.to/aws-heroes/multi-az-in-amazon-rds-and-how-it-...

How did that work out for people in the east region yesterday?

not good lol

yeah... no burgers or tacos https://www.theverge.com/2023/6/13/23759857/amazon-aws-down-...

that's WHY you need at least a multi-region HA database for your apps

No Roomba either. No multi-tasking housework while you work.

Perfectly for me.

(Maybe if I had used IAM auth for the database it wouldn't have?)

Or, uh, just self host.

What's the self-hosted Postgres HA story these days?

pg_auto_failover makes it an absolute breeze. I cannot understand how it's not mentioned in the article.

I've been runnning it for over 3 years with great success, and 0 downtime. Not to mention the extreme price difference compared to managed solutions - https://github.com/hapostgres/pg_auto_failover/discussions/6...

How are major version upgrades handled on PGA these days? Last time I used it (it's amazing), there wasn't a great path forward for minimal-downtime major version upgrades.

Same.

But major upgrades are only available once a year.

zero downtime for three years - congrats !

What about BU/DR? Is walg still the best?

I hosted postgres with cnpg operator in kubernetes. It does everything automagically: backups, wal backups and high-availability. Can't say nothing bad about it, it just worked for me. My only issue is that it releases versions too fast, I'd prefer some more relaxed schedule, but that's probably just me.

On self-hosted you can use the same as what cloud vendors are doing. Patroni or pg_auto_failover to manage single-primary + replicas. Maybe Neon to run an Aurora-like. CitusData... Azure did a lot around to get elasticity. Not easy to do the same. YugabyteDB on Kubernetes can be a cloud-native self-hosted solution

Which cloud vendors are using PGAF?

YugabyteDB can be self-hosted.

This is a completely different story. YugabyteDB is about transparent sharding and shard horizontal scalability. Every table is split into tablets. A tablet can be thought of as a shard but one still thinks of a table as a whole. Each tablet is replicated. Replicas can be placed in different regions, even in different computing environments. Want to run it across clouds or span the cloud and bare metal? Ignoring the cost and minding latency, why not. It’s still logical single database with whole tables. Want to put some rows in selected regions? No problem, still one logical table. Very different from master-replica architecture.