Member-only story
Kafka vs RabbitMQ vs NATS — I Cut the Network, 10k Messages Died
10k messages, gone in 10 seconds.
No error in the logs.
No alert.
The publisher was happy the whole time.
That was NATS.
Kafka lost 0 in the same test.
RabbitMQ lost 0 too.
And the honest part is that NATS did nothing wrong. It did exactly what it promises. I had just never read the promise.
The Test I Should Have Run First
We were picking a broker for an order pipeline, and like most teams, we started with throughput charts.
Every chart looked great.
None of them answered the only question that mattered to me: what happens to a message when the consumer disappears?
So I built the most boring test I could think of. 1 producer, 1 consumer, 1 broker, all in Docker Compose on my laptop.
The producer sent 1,000 messages a second, 512 bytes each, for 60 seconds. Every message carried a sequence number, so the consumer could count gaps and repeats.
Between the consumer and the broker I put Toxiproxy.
