問題文
A team claims their streaming pipeline delivers each record to the destination exactly once. Which combination does Structured Streaming actually require for that end-to-end guarantee?
選択肢
- A source that guarantees each record is delivered only once, so that the engine never sees a duplicate and therefore does not need the sink to have any particular property beyond being able to accept the writes.
- A transactional sink alone, because the engine deduplicates on the way in.
- The complete output mode, which rewrites the whole result and therefore cannot duplicate.
- A source that can be replayed by offset together with a sink that is idempotent, plus checkpointing of the processed offsets.