2 ms·
No mention of HA and last I checked Postgres still had no good solution for that... so I'm guessing they don't need HA and can tolerate downtime while they rest
by eikenberry 3y ago
No mention of HA and last I checked Postgres still had no good solution for that... so I'm guessing they don't need HA and can tolerate downtime while they restore the DB cluster?
- dijit 3y agoThe issue with “no good solution” is that sometimes things are inherently hard and certain technologies don’t permit lying. Good example is async in Rust; people don’t like it but mostly because async is hard. Postgres has excellent HA options, if you know what you’re trying to do with your data. CitusDB for data-warehousing storage, TimescaleDB for time-series data, and the traditional replication system for having HA (with a single write-primary) - which is the same method that Elasticsearch and ETCd are doing under the hood. Though in Elasticsearches case they do it by aggressively sharding their data set and splitting write masters across multiple nodes which has huge latency tradeoffs. Other multi-master HA systems have trade offs (or, lie about not having tradeoffs *cough* mongo *cough*).
- candiddevmike 3y agoPatroni or pg_auto_failover work well enough.
- viraptor 3y agoThey mentioned Aurora. That's AWS's solution to database clusters and it can do fancy HA.