Handle results and failures

Inspect the experiment status before using its measurements. Preserve failed runs as evidence rather than silently treating missing metrics as zero.

Load saved evidence

from pathlib import Path
import cloudai.api

experiment = cloudai.api.get_experiment(Path("results/your-experiment-directory"))
print(experiment.status)

Replace the path with the directory printed by execution. Use list_experiments(system) to discover saved experiments. See Read results and generate reports for the record structure and Python API for lookup errors.

Distinguish failures

Configuration and setup errors raise exceptions. Workload failures appear in the returned experiment’s status. Missing metrics need investigation even when a process completed. Store logs and the generated configuration so another process or person can diagnose the cause.

Cancellation

Pass on_start to receive the initial experiment and cancellation function before execution begins. The call remains synchronous; the callback is not an asynchronous execution API. Follow the contract in Python API when integrating with an application’s cancellation handling.