Really? Is there no acceptance that the state of PostreSQL replication is a huge hot mess? The fact that Uber couldn't figure it out should be a huge giant red flag about _both_ Uber's DBAs and _also_ the state of Postgres!
> "Uber had a lot of long-running transactions because [they were doing it wrong]"
pool-of-threads is a much better and much more defensive connection model. It's why NGINX beat out Apache - people LONG AGO realised process-on-request is terrible. How this can be shrugged off when pool-of-threads is easily one of MySQL's best features is a bit shocking. Has this guy ever been on-call for a web application before?
> Great for WordPress.
Lol! Bitter much?
Irrelevant
> Is this better or worse? Why? Uber doesn't say
It's worse. Obviously. Now you have introduced a race condition that only appears at load. Don't do this.
Fail fast and fail often.
> Great for WordPress.
Mocking, within the presentation, is not compelling.
> The consistent comparison of logical vs binary replication is maddening
There's no mention of community support around the logical replication. Statistically, nothing you look for will reference or confirm that the side feature added "years and years" ago, includes all the features from the binary form. Yes this is FUD-talk, but I'd be interested to know if Uber ran in to practical problems. When working with a complex software product, it's not uncommon for some less used product features to be broken, missing, or marked as "not going to fix" (e.g. Databricks, Silverlight) because they were less used and/or less important to the product aim.
This feels a lot like systemd debates (both in nomenclature and pragmatism).