Free engineering tool

Kafka Partition Capacity Planner

Q: How does the Kafka partition estimate work?

It divides peak ingress MB per second by tested throughput per partition multiplied by the utilization ceiling, rounds up, and applies a floor for consumer parallelism.

Q: Does this determine the right Kafka partition count by itself?

No. It does not model key skew, record size variation, broker disk or network limits, replication traffic, or consumer processing cost. Validate with representative benchmarks and operational limits.

Estimate a partition floor from peak ingress, benchmarked per-partition throughput, a utilization ceiling, and consumer parallelism.

Use representative measured throughput, not a vendor-independent default. Maintained by Buildopsy · Updated October 10, 2026.

Partition sizing scenario

The per-partition throughput input should come from a test matching payloads, acknowledgments, compression, and hardware.

Peak ingress (MB / second)

Tested throughput per partition (MB / second)

Maximum target utilization (percent)

Minimum parallel consumers

Replication factor

Estimated partition floor

higher of throughput and consumer floors

Runs locally · No data sent

Replicated copies

36 partition copies

partition count × replication factor

Aggregate throughput

120.0 MB/s aggregate

66.7% of tested capacity

Formula and assumptions

Throughput partitions = ceil(peak ingress ÷ (tested MB/s per partition × utilization ceiling)). Final count is at least the configured parallel consumer count. Replicated copies = partitions × replication factor.

This estimates a starting floor, not broker count or guaranteed capacity. It assumes evenly distributed keys and comparable partition performance. Hot keys, producer batching, message size, consumer work, storage, network, replication traffic, and rebalances can change the result. Do not create partitions solely to match current consumers; validate latency and throughput under realistic failure and recovery conditions.

Frequently asked questions

How does the Kafka partition estimate work?

It rounds up peak ingress divided by the tested per-partition rate at the chosen utilization ceiling, then enforces the requested consumer parallelism floor.

Does this determine the right Kafka partition count by itself?

No. Benchmark and account for key skew, broker and consumer bottlenecks, replication, and operational overhead before setting a production partition count.

Related calculators

All engineering calculators · Connection pool sizing · Queue backlog and drain time · Error budget runway