Execution Modes

execution_mode

CeleryLanguage supports three execution modes:

• local: run tasks in-process.
• distributed: dispatch via Celery broker.
• auto: choose behavior based on runtime context and fallbacks.

Local

Best for unit tests and quick feedback.

Properties:

• no external broker required
• deterministic call path
• emits job/subjob timeline events when journal is enabled

Distributed

Best for production-like async execution.

Properties:

• sends envelope to Celery task (lingo.execute)
• queue selection depends on routing_mode
• supports retry policy propagation and journaling

routing_mode

• shared: use a shared channel queue (default lingo.default).
• task: route by root task name; supports worker-specific task queues.

Phrase Local Placement

phrase(...).local(force=False) requests strict placement to one capable worker.

Behavior:

• if a worker with required tasks is registered, dispatch uses that worker's private queue
• if not found and force=False, dispatch raises
• if force=True, runtime may fall back according to mode/path

orchestration_mode

• simple: standard dispatch behavior.
• compiled: compiles workflow graph for optimized lowering paths.

Compiled mode notes:

• linear graphs can lower to Celery canvas chains
• fan-in shapes can lower to chords
• unsupported lowering shapes fall back to simple dispatch and emit a lowering event