What are common Apache Camel anti-patterns and performance pitfalls to avoid?

Knowing what NOT to do in Camel prevents subtle bugs and performance degradation. Here are the most frequently encountered pitfalls:

• Writing to exchange.getOut(): Creates a new Message and silently drops all headers set by prior processors. Always mutate exchange.getIn() instead.
• Long-running Processors on consumer threads: Blocking a consumer thread starves the thread pool. Move heavy work into a SEDA or thread-pool-backed pipeline.
• Unbounded in-memory aggregation: Aggregator without a completionSize or completionTimeout will grow the aggregation repository indefinitely until OOM. Always set a completion condition.
• Using MemoryIdempotentRepository in production: It is not persistent and not cluster-safe. Use JDBC or Infinispan repositories.
• Creating Endpoints inside a Processor: Calling context.getEndpoint() in a tight loop creates many endpoint objects. Cache endpoints or use ProducerTemplate with pre-created templates.
• Forgetting .end() after split() or choice(): Without end(), subsequent steps are placed inside the sub-route or branch, not the main pipeline.
• Swallowing exceptions silently: Setting handled(true) without routing to a DLQ or logging means errors disappear. Always log or route failed Exchanges.
• Overusing direct: for high-throughput flows: direct: is synchronous and blocks the calling thread. Use seda: or thread() for async fan-out.
• Large message bodies held in memory: Streaming data (large files, video) must use streaming() in split() and stream-aware data formats. Buffering multi-GB files causes OOM.
• Unmanaged thread pools: Calling .threads() without configuring pool size or using default settings in high-concurrency routes leads to thread exhaustion.

In production reviews, the most common issue is the getOut() header-loss bug, followed by unbounded aggregators. A Camel route code review checklist should verify: completion conditions on every aggregator, getIn()-only mutations, DLQ routing after error handlers, and streaming flags on file/byte-array splits.