Skip to content

startup-performance.md

Latest commit

 

History

History
39 lines (31 loc) · 1.64 KB

startup-performance.md

File metadata and controls

39 lines (31 loc) · 1.64 KB
Raw

Startup Performance

The autohand_sdk_tests executable guards wrapper-controlled startup with a deterministic in-process JSON-RPC fixture. Each metric uses five warmup iterations and 50 measured samples, reports median and p95 latency, and fails if any p95 reaches 50 ms.

  • publicImportMs: the first public autohand::initialize() call timed inside a fresh C++ process, excluding C++ runtime process boot. This idempotent hook performs the library's eager runtime initialization.
  • sdkStartReturnMs: elapsed time for the public AutohandSdk::start API, including its readiness getState request.
  • fixtureSpawnToFirstRpcMs: elapsed time to spawn the deterministic fixture and complete its first successful getState request.

Baseline captured on 2026-07-20:

Metric Median p95 Budget
publicImportMs 0.012083 ms 0.014959 ms < 50 ms
sdkStartReturnMs 3.140166 ms 3.888250 ms < 50 ms
fixtureSpawnToFirstRpcMs 3.450083 ms 4.102292 ms < 50 ms

Run the gate directly:

cmake --build build --target autohand_sdk_tests
./build/autohand_sdk_tests

The test emits one machine-readable JSON object with top-level language, budgetMs, metrics, and passed fields. Each metric contains samples, medianMs, p95Ms, maxMs, and passed. The fixed protocol is five warmups and 50 measured samples.

This gate isolates wrapper overhead. Real CLI startup may additionally perform provider authentication, network access, model loading, and other environment-specific readiness work; those live measurements are deliberately reported separately from the deterministic 50 ms gate.