What's happening
In internal/orchestrator/orchestrator.go, runTrial's stage loop returns immediately on any stage error:
if err != nil {
return failTrial(...)
}This applies to every stage (Prepare/Create/Start/WaitReady/Stop). If any of them fails, Cleanup never runs for that adapter.
Why it matters
This tool manages real containers/VMs/network namespaces on the host as root. A single failed trial in a benchmark sweep can leave those resources behind, and since sweeps run many trials, this accumulates across a run rather than surfacing as one isolated failure.
Suggested direction
Run Cleanup on a best-effort basis regardless of which stage failed (e.g. via defer or an explicit cleanup call in the failure path), so a failed trial doesn't skip teardown.
What's happening
In
internal/orchestrator/orchestrator.go,runTrial's stage loop returns immediately on any stage error:This applies to every stage (Prepare/Create/Start/WaitReady/Stop). If any of them fails,
Cleanupnever runs for that adapter.Why it matters
This tool manages real containers/VMs/network namespaces on the host as root. A single failed trial in a benchmark sweep can leave those resources behind, and since sweeps run many trials, this accumulates across a run rather than surfacing as one isolated failure.
Suggested direction
Run Cleanup on a best-effort basis regardless of which stage failed (e.g. via
deferor an explicit cleanup call in the failure path), so a failed trial doesn't skip teardown.