Frequent small inserts into ClickHouse create too many parts and eventually stall writes

Every INSERT creates at least one new part on disk, and background merges combine them. Inserting a few rows at a time outruns the merge scheduler, and once the number of active parts in a partition passes the configured threshold the server first delays inserts and then rejects them with a Too many parts exception.

The documented guidance is to batch inserts into large blocks, on the order of tens of thousands of rows, or roughly one insert per second per table, and to keep partitions coarse, typically monthly, since parts never merge across partitions. When the client cannot batch, enable server-side batching with async_insert, which buffers rows and flushes on a size or time threshold, and use wait_for_async_insert to control whether the client waits for that flush before treating the insert as durable.