Find queries with high memory usage ​

Rank terminal events by peak accounted memory to find requests that deserve inspection. This metric describes SQL memory for the request; it is not the process's full resident memory.

SELECT query_id, peak_memory_usage, query_duration_ms, query
FROM system.query_log
WHERE type = 'QueryFinish'
ORDER BY peak_memory_usage DESC
LIMIT 20
FORMAT json_each_row;

At terminal events, memory_usage and peak_memory_usage both report the peak accounted value. For a request still running, system.processes exposes current and peak values. Measurements remain available when max_memory_usage is zero.

Reduce the work first ​

Inspect wide projections, large grouped results, joins and unbounded sorts. Select only required columns, filter the input to a representative range, and consider whether the consumer actually needs every returned row. A smaller result can also reduce buffered response memory and response-byte pressure.

Repeat the same query on the same data before comparing a change. If a memory ceiling is necessary, choose it from measured workloads and remember that a saved database value cannot relax a stricter server ceiling. A higher limit alone does not explain why usage grew.

See memory limits, system.processes and EXPLAIN.