Real-Time Streaming Architectures Flashcards

A sliding window with a 10-minute size and 2-minute slide produces a new window every:

Answer: 2 minutes

The slide interval determines window emission frequency, so a new window starts every 2 minutes.

In Kafka, what does log compaction retain?

Answer: The latest value for each message key

Log compaction keeps at least the most recent value for every key, discarding older duplicates.

Which scenario most justifies exactly-once over at-least-once semantics?

Answer: Financial transaction aggregation where duplicates corrupt totals

Duplicate counting in financial aggregation produces wrong results, demanding exactly-once.

What is the main trade-off of a larger watermark delay (more lateness tolerance)?

Answer: Higher latency before results are emitted

Allowing more late data means windows stay open longer, delaying when results are produced.

In a Kafka Streams application, a KTable represents:

Answer: A changelog stream interpreted as an evolving table of latest values per key

A KTable models the latest state per key, updated as new records arrive on its changelog.

Which factor most directly limits the maximum parallelism of a Kafka topic's consumers in one group?

Answer: The number of partitions

Each partition is consumed by only one member, so partition count caps group parallelism.

What does an idempotent producer in Kafka prevent?

Answer: Duplicate messages caused by producer retries

An idempotent producer deduplicates retried sends so each message is written once per partition.