Author here. Yeah, pretty much everything in the article is an anti-pattern. However, people are doing these things, and the implementations are correct, and perform well enough for us. There are downsides, and dragons... beware!
What wasn't clear to me from the article is if Postgres can guarantee at-least-once semantics? I.e.: can it be guaranteed that messages are delivered to the consumer at least once given all kinds of failure situations?
Thanks. Reading the 'problems' section: a simple solution for "Resilience against crashing or hanging workers" would be having a TTL on the lock. If only...