# WorkerSpec

Source: https://www.tensorplay.cn/docs/generated/tensorplay.distributed.elastic.agent.server.WorkerSpec.html

```python
class tensorplay.distributed.elastic.agent.server.WorkerSpec(role: str, local_world_size: int, rdzv_handler: RendezvousHandler, fn: Callable | None = None, entrypoint: Callable | str | None = None, args: tuple = (), max_restarts: int = 3, monitor_interval: float = 0.1, master_port: int | None = None, master_addr: str | None = None, local_addr: str | None = None, event_log_handler: str = 'null', logs_specs: Any | None = None, start_method: str = 'spawn', redirects: Any = None, tee: Any = None, log_dir: str | None = None, virtual_local_rank: bool = False, numa_options: Any = None, duplicate_stdout_filters: list[str] | None = None, duplicate_stderr_filters: list[str] | None = None)
```

Blueprint of the worker group this agent manages.

Every node runs the same spec: the same role name, the same
local_world_size, and the same entrypoint semantics, so global rank
arithmetic across agents stays consistent.

```python
get_entrypoint_name() → str
```

Human-readable name of the entrypoint (module path or command).