roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Prepare for Apache Kafka interview questions on how the log stores, replicates and delivers events. Covers partitions and consumer groups, delivery semantics, Streams, Connect and operations.
845 of 845 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 845 new. 0 cut, outside your plan.
Most loops that bring up Kafka are not Kafka interviews. The subject arrives inside a backend, data or platform conversation, and each of those reaches into a different part of it. Backend teams ask about the clients: why a consumer that takes too long between polls is dropped from its group, what a retry does to ordering when several requests are in flight, how to publish an event and update a database row without losing one of the two. Data and streaming teams go further up: a Debezium connector turning table changes into events, a schema change that breaks a reader downstream, where exactly-once in Kafka Streams stops holding. Platform and SRE teams, who run the brokers, start from a failure: what `acks=all` promises when `min.insync.replicas` is one, how a follower falls out of the in-sync set, what an unclean leader election gives up to keep a partition writable. Two role roadmaps carry this subject inside a whole loop. The [Data Engineer roadmap](/roadmaps/data-engineer) is for the reader whose job is the pipeline rather than the broker: there Kafka is one stage of the loop, studied next to the storage and processing it feeds. The [Forward Deployed Engineer roadmap](/roadmaps/forward-deployed-engineer) suits a job that often meets Kafka as a customer's existing event bus, where the questions turn on connecting to it securely and reading from it without disturbing the teams already consuming it.
The [primer on the Kafka hub](/topics/cloud-kafka) already traces a record from producer to consumer and says which parts lean on which. The drilling happens on this page; the primer is where you settle it when two answers you gave about one setting turn out to contradict each other. The junior bar is the data model, explained without hesitation. [Topics, partitions and log storage](/topics/cloud-kafka-topics-partitions) comes first: what an offset addresses, why ordering holds inside a partition and not across them, what happens to existing keys when the partition count grows. Middle candidates are usually told apart on the [consumer side](/topics/cloud-kafka-consumers), where a rebalance that lands mid-batch, an offset committed before the work finished and lag that climbs while every consumer looks healthy supply the duplicate and delay stories interviewers build on. Senior candidates should reach [delivery semantics and transactions](/topics/cloud-kafka-delivery-semantics) early, because the exactly-once conversation draws on most of what sits beneath it: idempotent writes, fencing a zombie producer, the read-committed reader, and where the guarantee ends once a record leaves Kafka for a database. An application developer can leave the operator's side of the plan for a second pass. A platform candidate does the reverse and should expect the senior round there to be an incident told from the first signal checked: which metric moved, what it ruled out, what was changed and what it cost.
Most loops that bring up Kafka are not Kafka interviews. The subject arrives inside a backend, data or platform conversation, and each of those reaches into a different part of it. Backend teams ask about the clients: why a consumer that takes too long between polls is dropped from its group, what a retry does to ordering when several requests are in flight, how to publish an event and update a database row without losing one of the two. Data and streaming teams go further up: a Debezium connector turning table changes into events, a schema change that breaks a reader downstream, where exactly-once in Kafka Streams stops holding. Platform and SRE teams, who run the brokers, start from a failure: what acks=all promises when min.insync.replicas is one, how a follower falls out of the in-sync set, what an unclean leader election gives up to keep a partition writable.
Two role roadmaps carry this subject inside a whole loop. The Data Engineer roadmap is for the reader whose job is the pipeline rather than the broker: there Kafka is one stage of the loop, studied next to the storage and processing it feeds. The Forward Deployed Engineer roadmap suits a job that often meets Kafka as a customer's existing event bus, where the questions turn on connecting to it securely and reading from it without disturbing the teams already consuming it.
The primer on the Kafka hub already traces a record from producer to consumer and says which parts lean on which. The drilling happens on this page; the primer is where you settle it when two answers you gave about one setting turn out to contradict each other.
The junior bar is the data model, explained without hesitation. Topics, partitions and log storage comes first: what an offset addresses, why ordering holds inside a partition and not across them, what happens to existing keys when the partition count grows. Middle candidates are usually told apart on the consumer side, where a rebalance that lands mid-batch, an offset committed before the work finished and lag that climbs while every consumer looks healthy supply the duplicate and delay stories interviewers build on. Senior candidates should reach delivery semantics and transactions early, because the exactly-once conversation draws on most of what sits beneath it: idempotent writes, fencing a zombie producer, the read-committed reader, and where the guarantee ends once a record leaves Kafka for a database.
An application developer can leave the operator's side of the plan for a second pass. A platform candidate does the reverse and should expect the senior round there to be an incident told from the first signal checked: which metric moved, what it ruled out, what was changed and what it cost.