Ensuring Message Reliability in RocketMQ: Producer, Storage, and Consumer Guarantees

In a recent interview a candidate was asked whether a RocketMQ consumer that crashes would lose messages, which tests the candidate's understanding of how RocketMQ ensures message reliability.

The message lifecycle consists of three main phases: production, storage, and consumption. The article examines each phase to answer the interview question.

1. How the Producer Guarantees No Message Loss

RocketMQ provides three sending modes: synchronous, asynchronous, and one‑way. To avoid loss, the producer should:

Use synchronous sending so that the send method returns a success status only after the broker has stored the message.

If the send fails or returns a non‑success status, retry the operation.

Leverage transactional messages , which are designed to achieve zero loss.

2. How the Storage Layer Guarantees No Message Loss

Messages must be persisted to disk. RocketMQ supports two flush strategies:

Synchronous flush : the broker acknowledges only after the message is flushed to disk, guaranteeing no loss but reducing throughput.

Asynchronous flush : the broker acknowledges after the message is written to the page cache, improving performance but risking loss on power failure.

Broker nodes are typically deployed in a cluster with a master and slave . An ack is sent to the producer only after both master and slave have successfully written the message, implementing synchronous replication that prevents loss at the cost of throughput. Asynchronous replication returns an ack after the master writes, offering higher speed but possible loss.

3. How the Consumer Stage Guarantees No Message Loss

RocketMQ offers three consumption models:

At Least Once : each message is delivered at least once; duplicates may occur but loss is avoided.

At Most Once : each message is delivered at most once; loss may happen but duplicates do not.

Exactly Once : each message is processed exactly once, requiring transactional or idempotent handling.

Consumer progress is tracked via offsets. After processing, the consumer commits the offset to the broker, confirming successful consumption.

4. What Happens If the Consumer Crashes?

If the consumption mode is At Most Once , a crash can cause message loss. If the mode is At Least Once or Exactly Once , the message will not be lost.

Message retry : RocketMQ will retry undelivered messages within a configured interval; after exhausting retries, the message is marked as failed. Message persistence : Messages are stored on disk, so they survive consumer crashes. Offset tracking : Upon restart, the consumer resumes from the last committed offset.