I am always surprised at how infrequently logical replication is mentioned when people are trying to implement these types of systems. I've used Debezium with s

I am always surprised at how infrequently logical replication is mentioned when people are trying to implement these types of systems. I've used Debezium with some success in the past, but recently implemented my own application level reader to facilitate low latency event sourcing projections where I didn't want to involve Kafka. This seems to be quite a good solution in many scenarios that typically involve undesirable performance tradeoffs, as it doesn't introduce polling or much additional load on your DB, allows better isolation of the consumer workload, allows simple recovery if the consumer restarts or gets disconnected (e.g. you pick up at the LSN you left off and aren't worried that you missed a NOTIFY while offline), and watching status of the replication slot gives a since external metric to monitor of the health of the consumer.