Back-Pressure — Middle¶

At middle level, focus on this question:

How do push-based and pull-based flow control differ in how they achieve back-pressure?

Prerequisite: junior.md.

Pull-based: the consumer asks for more, on its own schedule¶

sequenceDiagram participant Consumer participant Producer Consumer->>Producer: give me the next item Producer-->>Consumer: item 1 Note over Consumer: processes at its own pace Consumer->>Producer: give me the next item (when ready) Producer-->>Consumer: item 2

In a pull-based model, the consumer explicitly requests each item (or batch) only when it's ready for more — back-pressure is implicit and automatic: the producer simply never sends faster than the consumer asks, because the consumer controls the pace entirely. Polling and traditional message queue "consume" APIs are pull-based by nature.

Push-based with explicit credit: the producer pushes, but only within a granted allowance¶

sequenceDiagram participant Consumer participant Producer Consumer->>Producer: I can handle 10 more\n(grant CREDIT) Producer->>Consumer: pushes up to 10 items Note over Consumer: processes, replenishes\ncredit as capacity frees up Consumer->>Producer: +5 more credit available now

This is the credit-based flow control mechanism from the Event-Driven Background Jobs professional page — the producer pushes proactively (lower latency per item than waiting for each individual pull request) but is still bounded by however much credit the consumer has explicitly granted, giving you push's latency benefit with pull's rate-control safety.

	Pull-based	Push-based + credit
Latency per item	Higher (must wait for next request)	Lower (producer sends proactively within its credit)
Complexity	Simpler	Requires credit tracking on both sides
Back-pressure	Automatic, implicit	Explicit, requires the credit protocol

🎓 Takeaway: pull-based flow control gets back-pressure "for free" at the cost of latency; push-based-with-credit recovers push's latency advantage while still providing explicit rate control — the credit mechanism is what prevents push-based systems from defaulting back to junior.md's uncontrolled-buffering problem.

Test yourself¶

Why is back-pressure "automatic" in a pull-based model, without any extra protocol needed?
Why does pure push (with no credit mechanism at all) recreate exactly the unbounded-buffering risk from junior.md?
Design a simple credit-based protocol for a producer pushing sensor readings to a consumer that can only reliably process 100 readings/sec.

Continue to senior.md.