Batch, Streaming, and Latency: 10 practice questions

Question 1 of 10

What makes a batch dataset bounded?

1. AIt continues receiving events for as long as the source runs
2. BIt emits each result before any later record can arrive
3. CIt has a finite set of records for the job to process
4. DIt requires a broker to retain an indefinite event history

Question 2 of 10

What does event time represent?

1. AWhen the event occurred in the producing domain
2. BWhen a consumer last updated its checkpoint directory
3. CWhen the current batch job was originally scheduled
4. DWhen a pipeline worker began processing the event

Question 3 of 10

What property makes a sink idempotent?

1. AThe sink changes event time to the current processing time
2. BThe sink discards each first delivery and stores only retries
3. CRepeating the same input leaves the same business result
4. DEvery retry creates a second independent business record

Question 4 of 10

Why does an unbounded stream need windows for many aggregations?

1. AA window creates a finite group from a source that keeps receiving records
2. BA window converts event time into a permanent batch schedule
3. CA window guarantees that records arrive in exact event-time order without late arrivals
4. DA window removes the need to retain state for every aggregation

Question 5 of 10

Which statement correctly separates latency from throughput?

1. AThroughput measures event occurrence time while latency measures source boundedness
2. BA pipeline may process many records per second while individual records still wait in a queue
3. CLow latency guarantees that the system handles the largest possible total volume
4. DHigh throughput guarantees that every event has millisecond end-to-end latency under any backlog

Question 6 of 10

What duplicate risk accompanies at-least-once delivery?

1. AThe consumer replaces event timestamps with batch-window boundaries
2. BA retry can apply an input whose earlier outcome was uncertain
3. CThe broker refuses to redeliver an input after a failed consumer
4. DThe pipeline drops every record that arrives after a watermark

Question 7 of 10

Which tradeoff generally distinguishes streaming from scheduled batch processing?

1. AStreaming reduces response delay but adds continuously operated state and recovery complexity
2. BStreaming increases response delay while eliminating producer and consumer services
3. CStreaming accepts only finite datasets but removes all monitoring needs
4. DStreaming processes historical batches only and cannot update results incrementally

Question 8 of 10

A fraud service must react within seconds, traffic arrives continuously, and occasional bursts must not lose events. Which design best fits?

1. AA streaming consumer without durable buffering because low latency prevents traffic bursts
2. BA weekly full-history batch with high throughput and no event-time handling
3. CA nightly batch that scans the complete day after the reaction window has passed
4. DA monitored streaming pipeline with durable buffering, recovery state, and scalable consumers

Question 9 of 10

Events can arrive ten minutes late, reports must group them by when they happened, and early results may be revised. Which clock and policy should drive aggregation?

1. AUse event-time windows with a watermark that tolerates the required lateness
2. BUse processing-time windows and assume arrival order matches occurrence order
3. CUse the batch schedule time and discard every timestamp carried by an event
4. DUse checkpoint write time as the business timestamp for each reported record

Question 10 of 10

A sink writes billing records, the streaming connector may retry a micro-batch, and lost bills are less acceptable than repeated delivery attempts. Which write design is safest?

1. AReplace each bill's event time with retry time and keep every repeated row
2. BUse at-least-once delivery with stable billing keys and idempotent merge behavior
3. CUse at-least-once delivery and insert a fresh billing row for every retry
4. DDisable redelivery and accept losing any record whose first outcome is uncertain