What it is
Every INSERT into a MergeTree table writes a new immutable part, and background merges combine neighbouring parts inside each partition. When parts arrive faster than merges can combine them, queries open more and more parts, inserts are delayed, and finally they fail with 'Too many parts'. This simulator shows parts as tiles, merges as they run, and compares ways of ingesting the same rows.
When to use it
Use it to see why tiny inserts hurt, what async inserts and bigger batches change, why adding merge threads only helps so much, why a high-cardinality partition key multiplies parts, and what FINAL costs on ReplacingMergeTree. Compare strategies on the same load, then replay one and change the load or slow the disk while it runs.
Gotchas & limits
This is a teaching model at the level of parts, not bytes: there are no columns, replicas, projections or mutations. The limits (1000 and 3000 parts), merge speeds and query costs are illustrative assumptions that were not checked against current ClickHouse documentation or a real server. The results show the mechanism, not a prediction for your cluster.