From the strictest CAP theorem definition, that's correct, it is not. But, it's pretty close. I know that in the database world, that's not a good answer, but in practice, it will deliver the vast majority of messages, so maybe that's good enough? We'll see.
We show that it's possible to come close without breaking the DB or the app, but I suspect, it's not quite yet at the level you'd expect from a _durable_ work queue, e.g., Kafka. Not going to replace that one anytime soon.
Comments
Doesn't this NOTIFY performance fix mean that it isn't transactional any more?
From the strictest CAP theorem definition, that's correct, it is not. But, it's pretty close. I know that in the database world, that's not a good answer, but in practice, it will deliver the vast majority of messages, so maybe that's good enough? We'll see.
We show that it's possible to come close without breaking the DB or the app, but I suspect, it's not quite yet at the level you'd expect from a _durable_ work queue, e.g., Kafka. Not going to replace that one anytime soon.
Please document this, if that's not the case already. Thank you for open sourcing this!