https://docs.ambar.cloud/ says the author's org is polling PostgreSQL (or MySQL) tables and producing records for JSON/HTTP consumers. As for me, I think a Kafka broker quorum is hard to beat for a fast and durable distributed ringbuffer.
We use Kafka under the hood. We stream instead of poll. We used to work at the Kafka team in AWS :)
Our thesis is that a big blocker is the PhD the whole team needs in Kafka. For example, if you want to set up an API similar to Ambar with tools like Kafka connect, the connectors have failure modes that will bite you once a year and bite you hard. Eg losing your changelog in MySQL, and having to start from scratch or risk losing ordering guarantees.
A: You have managed vendors that simplify Kafka, such as MSK, Confluent, and Redpanda. And other software like Pulsar.
But we believe the solution to the time sink exposed in the article lies one level of abstraction higher. In the case of data analytics, there are tools/companies such as Decodable/Streamkap/Airbyte that simplify your life as an engineer.
In the case of operational streaming, we (Ambar) are making a bet on the tried and tested outbox/inbox pattern as a replacement for producing directly into Kafka et al, and thus managing all of its quirks and complexities. That’s the alternative we offer, but of course there are other folks in this space.
Admittedly, we didn’t dive deep into alternatives in the post. But we did explain at the end that we’ll cover it in another post. I’ll add a link at the bottom later pointing to some alternatives. :)