added an enum for exit codes

Author: ngc92

src/discord-cluster-manager/consts.py

@@ -1,4 +1,4 @@
-from enum import Enum
+from enum import Enum, IntEnum
 from typing import Type
 
 
@@ -25,6 +25,22 @@ class ModalGPU(Enum):
     H100 = "H100"
 
 
+class ExitCode(IntEnum):
+    """
+    Exit codes for our runners. These are just the codes actively return,
+    others are possible (e.g., exiting due to segfault, permissions, signal, ...)
+    """
+
+    # program ran successfully
+    SUCCESS = 0
+    # a cuda API call failed
+    CUDA_FAIL = 110
+    # could not setup file descriptor for custom pipe
+    PIPE_FAILED = 111
+    # didn't crash, but tests failed
+    VALIDATE_FAIL = 112
+
+
 def combine_enums(enums: list[Type[Enum]], combined_name: str) -> Enum:
     combined_members = {}
     for enum in enums:

src/discord-cluster-manager/run_eval.py

@@ -5,7 +5,7 @@
 import time
 from typing import Optional
 
-from consts import CUDA_FLAGS
+from consts import CUDA_FLAGS, ExitCode
 
 
 @dataclasses.dataclass
@@ -141,10 +141,8 @@ def run_program(args: list[str]) -> RunResult:
         key, _, value = line.partition(":")
         result_dict[key.strip()] = value.strip()
 
-    # 0     everything was fine
-    # 112   program ran fine, but we detected a test failure
     return RunResult(
-        success=run_process.returncode == 0,
+        success=run_process.returncode == ExitCode.SUCCESS,
         command=_make_cmd(run_process.args),
         stdout=run_process.stdout,
         stderr=run_process.stderr,