
RabbitMQ de los fundamentos a producción, Parte 2: construir un servicio escalable
RabbitMQ de los fundamentos a producción, Parte 2: construir un servicio escalable Maté el...
Tag archive

RabbitMQ de los fundamentos a producción, Parte 2: construir un servicio escalable Maté el...

RabbitMQ from Fundamentals to Production, Part 2: Building a Scalable Service I killed the...

Before 143 dead-lettered messages go back to their consumer, send one and watch whether it dies again. How to replay a single message by hand, how long to wait given your retry policy, how to tell its return from a new failure, and when a passed canary still proves nothing.
"Do we need a queue or a log?" used to mean "RabbitMQ or Kafka?" That is no longer true. Apache Kafka...

Get messages, Move messages, Publish message and Purge, measured on RabbitMQ 4.3.6: which ack mode removes a message, which one costs a delivery on a quorum queue, where moved messages end up, and what to do before you click.
Small differences between two ZoomEye rounds are noise, while a large ZooKeeper gap comes from the query text rather than the internet.

Architect resilient, low-latency systems by combining Kafka's partition-based log streaming with RabbitMQ's smart-routing quorum queues under heavy re
A circuit breaker in every consumer turned one 15-second outage into 27 openings and ~2,000 dead letters. RabbitMQ fixed it with features it already had.

A replay that the broker confirmed is not a replay that worked. How to tag replayed messages so the ones that fail again can be found, what RabbitMQ 4 keeps and what it resets when a replayed message dies a second time, and how to read the result.

Where MassTransit puts a message when a consumer throws or nobody consumes its type, what the MT-Fault headers tell you, how retry and delayed redelivery interact with the error queue, and how to send the messages back to the endpoint without losing any.
We treated RabbitMQ as the record of our work. Then messages expired mid-job and nobody could answer "what failed?" The fix was an outbox table.

On RabbitMQ 4, a message dead-lettered from a quorum queue keeps its delivery count. With all defaults, a poison message arrives in a quorum DLQ with 21 deliveries against a limit of 20, and one crash, or even one peek, of whatever reads the DLQ loses it for good. Measured, explained, and three one-line fixes.