Optimizing ActiveMQ Message Queue Backlog: Removing Synchronization Locks and Tuning queuePrefetch

In a production environment the notification queue based on ActiveMQ suffered severe data backlog, causing transaction failures and requiring temporary merchant shutdowns.

Problem analysis identified that (1) massive message bursts exceed the consumer’s processing capacity, (2) the onMessage callback was declared synchronized , forcing serial processing, (3) removing the lock does not introduce concurrency safety issues because each consumer works on independent data, and (4) duplicate consumption is prevented by ACK mechanisms and database uniqueness constraints.

Phase 1 – Lock removal and prefetch tuning :

Test data of 15,000 messages with 10 ms processing time and consumer concurrency 5‑100 showed that with the lock in place only 15 consumers were active, taking 151 s (≈10 msg/s). After removing the synchronized keyword, processing time dropped to 13 s (≈1,150 msg/s). Further tuning the ActiveMQ queuePrefetch parameter from the default 1000 to 100 increased parallelism, reducing the total time to 6 s.

Code example of the original listener method:

public synchronized void onMessage(Message message, Session session)

Configuration to set queuePrefetch to 100:

tcp://localhost:61616?jms.prefetchPolicy.queuePrefetch=100

Phase 2 – Queue redesign :

Because the notification process is idempotent, the original single‑queue design mixed successful and repeatedly failing messages, causing blockage. A double‑queue approach was proposed: failed messages are moved to a second queue, isolating them from the main flow and improving overall throughput.

Phase 3 – MQ component re‑selection :

ActiveMQ’s throughput limitations suggest evaluating alternative brokers such as RabbitMQ, RocketMQ, or Kafka for higher performance in future migrations.

Conclusion :

By removing the synchronized lock (≈11× speedup) and adjusting queuePrefetch (≈2× further speedup), the overall consumption capacity increased over 30×, effectively resolving the backlog issue.