CCDAK Practice Test

CCDAK practice test: what's covered

• Producers
• Consumers
• Streams
• Schema Registry
• Connect
• EOS

Free CCDAK sample questions

1. Your team creates a topic 'orders' with 6 partitions and replication.factor=3 on a 5-broker cluster. A teammate asks how Kafka guarantees message ordering for this topic. What is the correct guarantee you should give them?

   • A. Ordering is guaranteed only within each individual partition, never across the whole topic
   • B. Ordering is guaranteed globally across all 6 partitions because they share one topic
   • C. Ordering is guaranteed only if the consumer reads with a single thread regardless of partitions
   • D. Ordering is guaranteed across the topic as long as replication.factor is at least 3
   Show answer

   Correct: A. Kafka guarantees offset ordering strictly within a single partition; records appended to a partition keep their relative order. There is no total order across partitions because partitions are written and consumed independently. Global ordering is wrong because the topic is split across 6 partitions. Single-threaded consumption does not create cross-partition order. Replication factor governs durability and fault tolerance, not ordering, so it is not the deciding factor.

2. During a broker failure drill, you notice a partition's ISR list shrank from 3 to 1 while min.insync.replicas=2 and the producer uses acks=all. What behavior should you expect for new produce requests to that partition?

   • A. Producers receive NotEnoughReplicas errors because the in-sync set is below min.insync.replicas
   • B. Producers continue successfully because the leader alone can still acknowledge writes
   • C. Producers automatically switch to acks=1 so writes keep flowing during the outage
   • D. Producers buffer indefinitely and the broker silently drops the oldest records
   Show answer

   Correct: A. With acks=all the leader only acknowledges once min.insync.replicas in-sync replicas have the record. When ISR drops to 1 but min.insync.replicas=2, the leader cannot satisfy the requirement and returns NotEnoughReplicas (a retriable error). Continuing to write would violate durability, so that option is wrong. Kafka never silently downgrades acks, and brokers do not silently drop produced records; the producer retries or eventually fails instead.

3. A platform engineer is enabling log compaction on a 'user-profiles' topic that stores the latest state per user keyed by userId. Which statements about compacted topics are correct? (Choose TWO)

   • A. Compaction retains at least the most recent value for each distinct message key
   • B. A record with a non-null key and a null value acts as a tombstone marking the key for deletion
   • C. Compaction guarantees every topic shrinks to exactly one segment after cleanup
   • D. Compaction requires all records to have null keys to be eligible for cleaning
   • E. Compaction reorders records by key so consumers always read keys alphabetically
   Show answer

   Correct: A and B. Log compaction keeps at least the latest value per key, which is ideal for state-like topics. A null-value record with a non-null key is a tombstone that, after delete.retention.ms, removes the key. Compaction does not collapse to one segment because the active segment is never compacted and multiple cleaned segments remain. Records need non-null keys to be compacted, not null keys, so that option is incorrect. Compaction never reorders records; offsets and order are preserved.

4. An operations engineer sets retention.ms=604800000 and segment.bytes=1073741824 on a high-throughput topic and asks why old data is not being deleted immediately after a week. What is the most accurate explanation to give?

   • A. Kafka deletes data at the segment level, so a segment is removed only after every record in it has aged past retention
   • B. Kafka deletes individual expired records one by one as soon as each crosses retention.ms
   • C. Retention only applies after the cluster restarts, so deletion waits for the next broker reboot
   • D. Retention.ms is ignored whenever segment.bytes is set, so only size-based deletion runs
   Show answer

   Correct: A. Kafka enforces retention by deleting whole log segments, not individual records. A segment is eligible for deletion only when its newest record is older than retention.ms (and the segment is closed), so an active or recent segment lingers until it rolls. Per-record deletion is wrong because Kafka never rewrites segments for time retention. Retention runs continuously, not only at restart. Time and size retention coexist; setting segment.bytes does not disable retention.ms.

5. A developer wants strong delivery guarantees and configures a producer with enable.idempotence=true but also explicitly sets acks=1 and max.in.flight.requests.per.connection=7. When the application starts, what happens?

   • A. The producer fails to start with a ConfigException because idempotence requires acks=all
   • B. The producer starts and silently ignores enable.idempotence since acks=1 disables it
   • C. The producer starts normally because idempotence is independent of acks settings
   • D. The producer starts but duplicates are still possible because acks=1 weakens idempotence
   Show answer

   Correct: A. The idempotent producer requires acks=all (and retries>0, max.in.flight<=5). If you explicitly set a conflicting value such as acks=1, the client throws a ConfigException at construction rather than silently downgrading. It does not silently ignore the flag, and it is not independent of acks because the idempotence guarantee depends on full-ISR acknowledgement. Saying duplicates are merely possible understates it; the producer will not even start with the conflicting configuration.

Last updated: 2026-10-19

CCDAK practice test FAQ