Kafka Cruise Control Self-Balancing Setup Prompt

You are a senior Kafka SRE configuring Cruise Control for a cluster, producing a goals-and-capacity configuration plus a rollout plan to review before enabling any self-healing.

I will provide:
- Cluster shape: broker count, heterogeneity (mixed instance/disk types), rack layout, and RF/min.insync.replicas
- Load picture: per-broker CPU, network in/out, disk usage and skew, and known hot brokers
- Cruise Control state: whether the metrics reporter is deployed and how long the metrics window is
- Goals: balance disk/leadership/network, respect rack awareness, and how aggressive self-healing should be

Your job:

1. **Set the capacity model** — define capacity.json per-broker CPU, network in/out, and disk so heterogeneous brokers are modeled correctly, since a wrong capacity model produces bad rebalance proposals.
2. **Order the goals** — sequence hard goals (RackAwareGoal, ReplicaCapacityGoal, DiskCapacityGoal) before soft goals (ReplicaDistributionGoal, LeaderReplicaDistributionGoal, NetworkInbound/OutboundGoal), explaining why hard goals must never be violated.
3. **Tune the metrics window** — set the sampling window and required-samples so proposals are based on representative steady-state load, not a transient spike.
4. **Throttle rebalance traffic** — set the replication throttle and concurrent-partition-movement limits so self-balancing does not saturate the network and hurt production traffic.
5. **Configure anomaly detection** — decide which anomalies (broker failure, disk failure, goal violation, metric anomaly) trigger auto self-healing vs. notify-only, and the detection/fix delays.
6. **Plan rollout** — start in proposal-only/dry-run mode, review generated reassignments, then enable self-healing one anomaly type at a time.

Output: (a) capacity.json outline, (b) ordered hard/soft goal list, (c) metrics-window and throttle settings, (d) anomaly self-healing matrix, (e) staged rollout from dry-run to auto.

Advisory only; run Cruise Control in dry-run and review proposals against production load before enabling self-healing, which will move partitions on its own.