Skip to content

Back-Pressure — Junior

At junior level, focus on this question:

Why are both "buffer everything forever" and "silently drop what doesn't fit" bad default responses to a slow consumer?

Unbounded buffering: delays the problem, doesn't solve it

flowchart LR Fast["Fast producer"] --> Buffer["Unbounded buffer\n(keeps growing)"] Buffer --> Slow["Slow consumer"] Buffer -.grows forever if\nconsumer stays slow.-> Crash["Eventually: out of\nmemory, or unacceptable\nprocessing delay"]

If a producer is faster than a consumer and nothing limits the buffer between them, the buffer just keeps growing — this is the exact unbounded-queue risk covered in the Queue-Based Load Leveling reliability pattern: fine for a temporary burst, catastrophic for a sustained mismatch.

Silent dropping: loses data with no signal

flowchart LR Fast["Fast producer"] --> FullBuffer["Buffer full"] FullBuffer -.new items silently\ndiscarded.-> Lost["Data LOST, with\nNO signal to anyone\nthat this happened"]

The opposite extreme — silently discarding items once a buffer is full — avoids unbounded memory growth but loses data invisibly, with no feedback to the producer or any operator that something is being dropped. This is often worse than either extreme: you get neither the safety of "nothing is lost" nor the visibility of "we know exactly how overloaded we are."

What back-pressure actually does instead

flowchart LR Consumer["Consumer signals its\nACTUAL current capacity"] --> Producer["Producer explicitly\nSLOWS DOWN or PAUSES\nbased on that signal"]

Back-pressure means the consumer's actual capacity is explicitly communicated back to the producer, and the producer adjusts its rate in response — rather than either party guessing, buffering blindly, or dropping silently.

🎓 Takeaway: the goal of back-pressure is to make "the consumer can't keep up" an explicit, visible signal that changes the producer's behavior, rather than a silent problem that manifests later as memory exhaustion or invisible data loss.

Test yourself

  1. Why does unbounded buffering only delay, rather than solve, a sustained producer/consumer rate mismatch?
  2. Why is silently dropping data often worse than either buffering or an explicit backpressure signal?
  3. What would an explicit backpressure signal actually look like in a simple example — what information does the consumer need to communicate?

Continue to middle.md.