Skip to content

Returning Results — Junior

At junior level, focus on this question:

Which background jobs actually need to report a result back, and which are genuinely "fire and forget"?

The caller has already moved on

sequenceDiagram participant Caller participant Queue participant Worker Caller->>Queue: submit job Queue-->>Caller: "accepted" (job ID) Note over Caller: Caller's request/function call\nRETURNS HERE - it does not wait\nfor the job to finish Queue->>Worker: deliver job (later) Worker->>Worker: processes... (could take\nseconds or hours)

Unlike a normal function call (result = do_something(), which blocks until do_something returns a value), submitting a background job returns immediately, typically with just a job ID acknowledging it was accepted — the actual processing, and any result it produces, happens later, disconnected from the original call.

Two categories

Category Example Does the caller need the result?
Fire-and-forget "Log this analytics event," "send this metric" No — the job's side effect matters, but nothing waits on or reads a return value
Result-needed "Generate this PDF report," "run this ML inference," "process this payment" Yes — something (the user, another service) needs to know the outcome, and often the actual output data
flowchart LR Job[Background job] --> Q{Does anything\nneed the result?} Q -->|no| FireForget["Fire-and-forget:\nno result mechanism needed"] Q -->|yes| Needed["Result-needed:\nrequires an explicit\nretrieval mechanism"]

🎓 Takeaway: don't build result-retrieval infrastructure for jobs that don't need it — it's real, avoidable complexity and cost (storage, polling load, callback infrastructure). But for jobs where something genuinely needs the outcome, you must design an explicit mechanism, because there's no implicit "return value" the way a synchronous function call has.

Test yourself

  1. Classify these as fire-and-forget or result-needed: "send a welcome email," "generate a user's annual tax summary PDF," "increment a page view counter," "process a video upload into multiple resolutions."
  2. Why is "the job ran, I assume it worked" an insufficient design for a result-needed job?
  3. What real cost would you be adding unnecessarily if you built full result-polling infrastructure for a purely fire-and-forget job?

Continue to middle.md.