Kafka Partition Calculator

Size Kafka partitions from produce throughput, consumer parallelism, and retention — with storage and per-broker sanity checks.

Estimate partitions from throughput, consumer parallelism, and retention storage — then sanity-check against broker capacity.

Target produce throughput MB/s Per-partition produce capacity MB/s Consumer instances (desired parallelism) Peak messages / second Avg message size bytes Replication factor Retention days Broker count for skew check

Estimates only — real hot partitions, key cardinality, and consumer lag can require more partitions. Prefer increasing partitions gradually; shrinking is expensive.

How Partition Count Is Chosen

Kafka scales consumers in a group roughly one partition per active consumer for a topic. You also need enough partitions so aggregate produce/consume throughput fits within what each partition can sustain.

This calculator takes the maximum of:

Throughput-based partitions = target MB/s ÷ per-partition MB/s capacity
Parallelism-based partitions ≈ number of consumers you want running in parallel
Message-volume check using peak messages/sec × average size

Practical Tips

Prefer increasing partitions when scaling out; decreasing later requires recreating the topic.
More partitions improve parallelism but raise open file handles, rebalance time, and end-to-end latency under some workloads.
Watch key cardinality — a hot key funnels traffic into one partition regardless of count.
Round toward a multiple of broker count when you care about even replica placement.

Typical Per-Partition Capacity

Use your own benchmarked numbers when you have them. Ballpark starting points for planning are often on the order of 5–20 MB/s per partition depending on disk, batching, compression, and follower fetch load — measure on your hardware.