Performance Guidance

Status: Available in ETLantic 0.20.0 framing. ETLantic publishes no capacity, throughput, or latency claims for production sizing.

Recorded smoke baselines (historical, ETLantic 0.10.0) prove only that a small harness completed in one environment. They are not production evidence. See Performance Baselines.

Run the local harness

uv sync --group dataframes
uv run python benchmarks/dataframe_scale.py polars
uv run python benchmarks/dataframe_scale.py pandas

Record commit, Python and dependency versions, hardware, OS, dataset shape, warm-ups, samples, median, p95, and raw results. Do not compare unlike environments as controlled benchmarks.

Measure your workload

Benchmark representative graphs, data shapes, I/O, concurrency, and failure paths on the engines you will deploy. Plans are data-only coordination artifacts: construction does not read source rows or execute transformations. Execution cost belongs primarily to selected backends.

Separate validation/planning overhead from backend execution and I/O. See Benchmark Design.

Evaluator note

Until representative 0.18+ baselines are published, treat performance as adopter-measured. Do not infer warehouse throughput from framework smoke timings.