Flow Skill

Use Flow only when concurrency over larger collections materially helps. Prefer Enum or Stream for small, simple, or latency-sensitive paths.

Good Uses

• Large bounded collections
• CPU-heavy transformations
• Parallel aggregation by partition key

Guardrails

• Call Flow.partition/2 only when related items must land on the same reducer.
• Avoid partitioning for embarrassingly parallel work.
• Keep reducers stateful and mapping stages stateless.
• Validate memory and throughput tradeoffs before adopting Flow in hot paths.

Typical Shape

source
|> Flow.from_enumerable()
|> Flow.map(&transform/1)
|> Flow.partition(key: &partition_key/1)
|> Flow.reduce(fn -> initial_state() end, &accumulate/2)
|> Enum.to_list()

If the pipeline becomes a long-lived or back-pressure-sensitive stream, prefer GenStage instead.