Streaming Capacity & Partition Planner

Act as a senior streaming-platform engineer. Build a capacity and partition plan that remains safe during traffic spikes, broker loss, rebalances, and downstream slowdowns. Broker and processor: [Kafka/Kinesis/Pulsar + Flink/Spark/Kafka Streams] Average and peak throughput: [events/sec and MB/sec] Event size distribution: [p50/p95/p99] Key cardinality and skew: [details] Ordering requirement: [per key / none / global] Processing cost per event: [if known] Latency SLO: [p95/p99] Retention and replay window: [duration] Replication and availability target: [requirements] Expected growth over 12–24 months: [percentage] Provide: 1. Partition/shard count calculation with formulas and assumptions 2. Producer, broker, consumer, network, and storage capacity estimates 3. Keying strategy and hot-partition analysis 4. Consumer parallelism, checkpointing, and autoscaling policy 5. Headroom required for N-1 failure and peak replay 6. Backpressure and overload-shedding behavior 7. Rebalance, deployment, and scaling risks 8. Load-test plan with success criteria 9. Operational metrics and alert thresholds for lag, skew, throughput, and saturation 10. Revisit triggers that indicate repartitioning is required Show calculations. Distinguish measured inputs from estimates and include a sensitivity analysis for the largest unknowns.