warn on unknown runopt passed via command-line

Summary:
When you pass an invalid scheduler arg via the .torchxconfig file, we get a warning about it.

This change adds a similar warning when the scheduler arg is passed via a command-line argument.

(background: I spent a few hours yesterday debugging a torchx command for a job until spotting a typo in one of the arguments)

Differential Revision: D70562266

Author: Daniel Ohayon

torchx/specs/api.py

@@ -63,6 +63,9 @@
 #     (most recent call first):
 _EMBEDDED_ERROR_MESSAGE_RE: Pattern[str] = re.compile(r"(?P<msg>.+)\nException.*")
 
+YELLOW_BOLD = "\033[1;33m"
+RESET = "\033[0m"
+
 
 # ========================================
 # ==== Distributed AppDef API =======
@@ -937,6 +940,10 @@ def _cast_to_type(value: str, opt_type: Type[CfgVal]) -> CfgVal:
             runopt_ = self.get(key)
             if runopt_:
                 cfg[key] = _cast_to_type(val, runopt_.opt_type)
+            else:
+                logger.warning(
+                    f"{YELLOW_BOLD}Unknown run option passed to scheduler: {key}={val}{RESET}"
+                )
         return cfg
 
     def cfg_from_json_repr(self, json_repr: str) -> Dict[str, CfgVal]: