Topic 17.4
Senior and Staff Questions (51–65)
In one line
Fifteen failure-driven questions: producer timeouts after acceptance, crashes before commit, growing lag, a partition with half the traffic, 15-minute batches, 95% full disks, shrinking ISR, broker death during leadership, AZ survival, global ordering, per-customer ordering, a 30% customer, exactly-once payments, atomic DB + Kafka writes, and ZooKeeper-to-KRaft migration.
Think of it like this
A senior pilot's check ride, where the examiner fails an engine at the worst moment and watches how you reason.
Key ideas
- 01
Answer shape: precise mechanism, what's lost or duplicated, how you'd detect it, and the configuration or design that prevents it.
Explain it without notes
51. A producer receives a timeout after the broker accepted the record. What happens?
52. A consumer processes a message successfully but crashes before committing the offset. What happens?
53. Consumer lag is growing continuously. How do you diagnose it?
54. One partition is receiving 50% of all traffic. What do you do?
55. One consumer takes 15 minutes to process a batch. What happens?
56. A broker's disk is 95% full. What do you do?
57. The ISR keeps shrinking. What do you investigate?
58. A broker dies while leading partitions. What happens?
59. You need to survive one AZ failure. How do you configure Kafka?
60. You need global ordering. Can Kafka provide it at scale?
61. You need strict ordering per customer. How do you partition?
62. One customer generates 30% of traffic. How do you solve the hot partition?
63. You need exactly-once payment processing. What does Kafka guarantee and what must the application guarantee?
64. A PostgreSQL update and a Kafka publish must be atomic. How do you solve it?
65. How would you migrate from a ZooKeeper-era cluster to KRaft?
Done when you can
I can answer all 15 staff-level questions with mechanisms, detection and prevention.