Measure application workloads ​

Measure an integration change with a workload that represents the application's task. A useful result includes correct rows, the complete request path and repeated measurements under stated conditions.

Define the operation ​

State the dataset, schema, query, output format, interface and connection lifetime. Decide whether connection setup belongs in the measurement. Reusing one native connection and making a new HTTP request exercise different session boundaries; compare them only when that distinction matches the application's question.

Run correctness checks before collecting timings. Use fixed inputs and explicit ordering for result comparisons. Separate the first run from repeated runs and report the individual observations or their spread rather than only the fastest number.

Read public response metadata ​

The HTTP interface returns X-Coordinator-Summary with fields including elapsed_ms and result_rows. Compare that server-reported duration with the application's own end-to-end duration. Record X-Coordinator-Result-Truncated when present so a capped result is not mistaken for a complete export.

Keep host details and simultaneous application activity with the report. Avoid attributing a timing difference to an unobserved implementation detail. For ingest measurements, use the separate compressed-input benchmark guide to control file preparation and row validation.