load-testing

You can find the maximum QPS for your Databricks App by running controlled Locust-based streaming load tests that ramp users to saturation, measuring TTFT and latency until performance degrades. The skill automates this orchestration and saves results to a run-specific dashboard.

To benchmark Databricks App performance across compute sizes, you provide the deployment URLs, compute sizes, and worker counts. The skill runs a test matrix, measuring peak QPS, latency percentiles, and failure rates, then generates an interactive HTML dashboard to compare the configurations.

Yes, you can load test SSE streaming endpoints and measure time to first token. The skill's Locust-based harness includes SSE stream parsing and tracks TTFT alongside standard latency metrics during the ramp-to-saturation performance benchmarking process.

Yes, you need M2M OAuth tokens to authenticate load tests on your Databricks App. The skill includes M2M OAuth token handling within the Locust test scripts to ensure your streaming requests are properly authorized during the performance benchmarking.

Yes, you can benchmark app throughput without real LLM calls by enabling the optional mocked LLM throughput mode. This allows you to measure the infrastructure's maximum sustainable QPS and streaming capacity independently of downstream model latency.