compute:run
compute:run
Section titled “compute:run”Run governed local execution or optional HTTP assignment polling
Options:
| Flag | Description | Required |
|---|---|---|
--experimental | Deprecated compatibility flag; local execution is stable and no longer requires —experimental | No |
--workspace-id <id> | Workspace id to attribute measured node energy to. When set, the runner auto-discovers observe-only RAPL/NVML sensors and emits measured joules beside each assignment. Omit to run without node energy metering. | No |
--workspace-root <path> | Workspace root fallback for local tool execution (overridden by assignment trigger_context) | No |
--drain-file <path> | Drain marker: when this file exists (or on SIGTERM/SIGINT) the runner finishes the assignment in progress, claims no more work and exits 0. Written by compute:service drain. | No |
--control-file <path> | Optional local supervisor halt JSON (assignment_id and pending_control_action: kill) | No |
--assignment-file <path> | Local assignment JSON; scope comes from the persisted WU claim | No |
--max-iterations <n> | Cap iterations (default: unbounded) | No |
--runner-dir <path> | Runner directory (default: ~/.lumenflow) | No |
--poll-interval <ms> | No | |
--session-token-env <name> | Environment variable containing the workspace API token for session registration | No |
--register-session | Register or refresh the session through /api/v1/sessions before polling. OPT-IN: the control plane currently resolves registration and claim to different session keys, so an unsolicited registration can shadow the runtime-owned session and make every later claim fail. Pass this only when the control plane you target resolves both to one key; otherwise the runner claims with its computed capability set and a narrower registered session shows up as 403 capability_mismatch. | No |
--session-id <id> | Session id for the claim probe | No |
--capabilities <csv> | Comma-separated capability set (default: empty) | No |