Instruction file imported from iPzard/coding-agents (.github/instructions/architecture-scalability/backpressure.instructions.md). Copyright stays with the author.

Backpressure

Cite as: [Backpressure]

Principle

If a producer can outrun a consumer, the system must drop, buffer, or block. Unbounded in-memory queues are bombs — they look fine until traffic spikes and the process runs out of memory.

Red flags in code

• Unbounded in-memory queue / buffer / timer-driven accumulator with no max size.
• A producer with no flow-control mechanism.
• A consumer that processes asynchronously without a bounded work queue.

Apply directly (mechanical, unambiguous)

• Add a maxSize / capacity to an unbounded queue when the policy is clear (drop, block, or reject).

Propose only (requires judgment)

• Choosing the right backpressure policy (drop oldest, reject newest, block producer).