Self-Assessment Quiz: Event Streaming with Apache Kafka

Twenty questions. Aim for 16 or more. On the streaming path, treat 18 as the bar.

Question 1

A Kafka topic is, as a data structure:

• A. A queue
• B. An append-only log, partitioned, read by position, not consumed by reading
• C. A key-value store
• D. A distributed hash table

Question 2

Which property does a queue NOT have?

• A. Ordering
• B. Durability
• C. Replay
• D. Delivery

Question 3

The partition key does three jobs at once. Which is NOT one of them?

• A. Determines ordering — same key, same partition
• B. Spreads load across partitions
• C. Determines retention
• D. Determines skew

Question 4

A null key:

• A. Fails the produce
• B. Round-robins, maximizing distribution and giving up ordering entirely
• C. Uses partition 0
• D. Uses the message timestamp

Question 5

Increasing a topic's partition count:

• A. Is a free capacity change
• B. Breaks per-key ordering across the change, permanently, for straddling messages
• C. Rewrites existing messages
• D. Requires a full consumer restart only

Question 6

acks=all without min.insync.replicas set can degrade to:

• A. acks=0
• B. acks=1, when the in-sync set falls to one replica
• C. No acknowledgment at all
• D. It cannot degrade

Question 7

The most common producer bug is:

• A. Wrong serialization
• B. No delivery callback, so failures are silently dropped
• C. Too large a batch size
• D. Missing compression

Question 8

enable.auto.commit=True:

• A. Is safe and recommended
• B. Commits on a timer with no knowledge of whether processing succeeded — silently at-most-once
• C. Commits only after a successful poll
• D. Is required for consumer groups

Question 9

Which liveness mechanism causes most rebalance problems?

• A. session.timeout.ms — the background heartbeat
• B. max.poll.interval.ms — the poll loop
• C. heartbeat.interval.ms
• D. connections.max.idle.ms

Question 10

A rebalance storm is self-sustaining because:

• A. Kafka retries indefinitely
• B. Reassignment gives the surviving consumers more work, so they also exceed the interval
• C. Offsets are lost
• D. The coordinator fails over

Question 11

The first fix to try for a rebalance storm is:

• A. Add consumers
• B. Raise max.poll.interval.ms
• C. Reduce max.poll.records
• D. Increase partition count

Question 12

Which does NOT help a rebalance storm caused by per-batch slowness?

• A. Reducing max.poll.records
• B. Processing asynchronously with pause/resume
• C. Adding consumers
• D. Raising max.poll.interval.ms

Question 13

group.instance.id enables static membership, which:

• A. Prevents all rebalances
• B. Lets a consumer restarting within the session timeout rejoin with its existing assignment
• C. Pins partitions permanently
• D. Disables the group coordinator

Question 14

Compaction keeps:

• A. Messages for a time limit
• B. The most recent message per key, forever
• C. All messages
• D. Only messages with non-null keys

Question 15

What makes a compacted topic reconstructible into a table?

• A. Its ordering
• B. Replaying from the beginning gives the current state of every key
• C. Its replication factor
• D. Its retention

Question 16

Retention is described primarily as:

• A. A cost decision
• B. A recovery-window decision — how long a consumer can be broken before recovery gets harder
• C. A compliance decision
• D. A performance decision

Question 17

The idempotency key Kestrel's bronze writer uses is:

• A. The event id
• B. The session id
• C. (topic, partition, offset) — unique by construction
• D. A content hash

Question 18

Little's Law gave 8.7 handlers and the multipliers suggested 26. The topic has 12 because:

• A. The arithmetic was wrong
• B. The consumer parallelizes internally, 12 has convenient divisors, and the peak lasts minutes
• C. 26 partitions exceeded broker limits
• D. Ordering required fewer partitions

Question 19

Among crash, skip, and dead-letter, the chapter recommends:

• A. Crash — it is loud
• B. Skip — it maintains throughput
• C. Dead-letter, and only if someone reads it
• D. It depends on the message

Question 20

A DLQ's retention should be:

• A. Shorter than the source topic's
• B. The same as the source topic's
• C. Longer than the source topic's — its contents are what you have not dealt with yet
• D. Unlimited

Answer Key

1. B — §15.1. Everything else in the chapter is a consequence.

2. C — §15.1. Reading removes the message, so a bug is unrecoverable and a new consumer cannot read history. Replay alone justifies Kafka for most data engineering.

3. C — §15.2. Retention is a topic setting, unrelated to the key.

4. B — §15.2. Appropriate for metrics; wrong for anything with a per-entity sequence.

5. B — §15.2. It is a decision with an ordering consequence, not a capacity knob.

6. B — §15.3. acks=all with an in-sync set of one is acks=1 wearing a different name. Set the minimum explicitly.

7. B — §15.3. produce() returns before the message is sent; the symptom is missing data with no error anywhere.

8. B — §15.4. And it is the default.

9. B — §15.5. Your process is alive and heartbeating; it is simply slow.

10. B — §15.5. The system's response to slowness makes it slower, and it does not recover on its own.

11. C — §15.5. The fastest fix and usually sufficient.

12. C — §15.5. More consumers each still take too long, and you have added rebalance churn.

13. B — §15.5. Removes deploy-time churn for one line of configuration.

14. B — §15.6.

15. B — §15.6. Which is why CDC topics are compacted, and why the tombstone matters.

16. B — §15.6, 💸 callout. The same question as max_slot_wal_keep_size and vacuum retention, in three different systems.

17. C — §15.7. Unique by construction and requires no cooperation from the payload.

18. B — §15.8. The arithmetic gives you a floor and a shape, not an answer.

19. C — §15.10. Crash stops everything after the bad message; skip is silent data loss.

20. C — §15.10, ⚠️ callout. The reverse — the default — is how 8 months of diverted messages aged out.

Topic map