Queuing

Canonical guidance

• make queue size part of the design
• decide what happens when producers outrun consumers
• keep shutdown and draining semantics explicit

Use when

• bounded work queues
• burst absorption
• handoff between producer and worker stages

Avoid

• unbounded backlog by accident
• raising queue sizes until tests stop hanging
• using a queue where direct backpressure would be clearer

Preferred pattern

• pair queue capacity with a documented overload policy

Anti-pattern

• a buffered channel with a magic size and no explanation

Explanation: This is tempting because a buffer can stop blocking temporarily, but it often only delays overload visibility.

Why

• queues control resource pressure and failure behavior