CCAAK study plan

Install a local single-node Kafka (KRaft mode) and a 3-broker cluster using containers or VMs, then verify each node's broker.id and log.dirs are distinct.

Diagram the KRaft controller quorum and explain in your own notes how controller and broker roles can be separated versus combined.

Map every server.properties setting you changed to the purpose it serves (listeners, advertised.listeners, log retention, replication defaults).

Create a topic with explicit partition count and replication factor, then inspect it with kafka-topics.sh --describe and confirm leader and ISR placement.

Configure replication and durability knobs (min.insync.replicas, acks, unclean.leader.election.enable, replica.lag.time.max.ms) and justify each value for a given durability target.

Tune log retention and compaction settings and demonstrate the difference by producing records to a compacted topic and observing key overwrites after restart.

Set up quotas (producer_byte_rate, consumer_byte_rate) for a client ID and verify throttling takes effect using broker metrics or logs.

Configure a listener scheme with separate internal and external advertised listeners and connect one client through each.

Provision a multi-node cluster from scratch using automation (Ansible, shell scripts, or a container orchestration manifest) so a rebuild takes under 30 minutes.

Configure rack awareness with broker.rack and verify partition replica distribution across racks.

Plan disk layout and sizing: estimate log segment growth for a stated throughput and set log.retention.bytes or log.retention.hours accordingly.

Secure inter-broker and client traffic with SASL/SCRAM or mTLS, and confirm an unauthenticated client is rejected.

Execute a rolling restart of the cluster without dropping producer or consumer availability, and record the observed impact on under-replicated partitions.

Configure JMX or the metrics reporter and scrape broker metrics into Prometheus, then build a dashboard panel for under-replicated partitions and active controller count.

Write alert rules for offline partitions, under-min-ISR partitions, and disk usage above a threshold.

Practice a broker failure by killing one node and document how leadership moved and recovered.

Enumerate the key broker, producer, consumer, and controller metrics and state the failure each one would reveal.

Diagnose a deliberately induced consumer lag scenario using consumer group offsets and lag metrics, then identify the bottleneck.

Interpret Kafka server logs for common warnings (ISR shrink/expand, leader election, request queue saturation) and classify each as benign or actionable.

Set log level and log rotation for broker logs so that a defined retention window is preserved without filling disk.

Build a troubleshooting decision tree covering producer timeouts, consumer rebalance storms, and unavailable partitions.

Reproduce a rebalance storm by misconfiguring session timeouts or max.poll.interval.ms, then fix it and confirm group stability.

Perform a partition reassignment with kafka-reassign-partitions.sh to rebalance leadership and verify the result.

Practice recovering from a full disk on one broker by adjusting retention and moving partitions.

Execute a controlled upgrade of the cluster to a newer Kafka version and document the ordering constraints for protocol and log format versions.

Plan and perform a capacity expansion by adding a broker and redistributing partitions with minimal data movement.

Test backup and recovery of topic data or metadata and verify restored topics match expected offsets and configuration.

Write an incident postmortem template and fill it in for one simulated outage you caused during practice.

Re-derive configuration trade-offs from memory for replication, durability, and retention without consulting notes.

Complete a timed mixed-domain practice set and review every missed item back to its source concept.

Re-run your deployment automation from a clean environment to confirm reproducibility.

Consolidate your notes into a single reference sheet capped at two pages.