Merge pull request #31 from amd/alex_rocm_fix

Rocm plugin update to accommodate for changes in 7.0

Author: alexandraBara

nodescraper/plugins/inband/rocm/rocm_collector.py

@@ -43,19 +43,34 @@ def collect_data(self, args=None) -> tuple[TaskResult, RocmDataModel | None]:
         Returns:
             tuple[TaskResult, RocmDataModel | None]: tuple containing the task result and ROCm data model if available.
         """
-        res = self._run_sut_cmd("cat /opt/rocm/.info/version")
-        if res.exit_code == 0:
-            rocm_data = RocmDataModel(rocm_version=res.stdout)
+        version_paths = [
+            "/opt/rocm/.info/version-rocm",
+            "/opt/rocm/.info/version",
+        ]
+
+        rocm_data = None
+        for path in version_paths:
+            res = self._run_sut_cmd(f"grep . {path}")
+            if res.exit_code == 0:
+                rocm_data = RocmDataModel(rocm_version=res.stdout)
+                self._log_event(
+                    category="ROCM_VERSION_READ",
+                    description="ROCm version data collected",
+                    data=rocm_data.model_dump(),
+                    priority=EventPriority.INFO,
+                )
+                self.result.message = f"ROCm: {rocm_data.model_dump()}"
+                self.result.status = ExecutionStatus.OK
+                break
+        else:
             self._log_event(
-                category="ROCM_VERSION_READ",
-                description="ROCm version data collected",
-                data=rocm_data.model_dump(),
-                priority=EventPriority.INFO,
+                category=EventCategory.OS,
+                description=f"Unable to read ROCm version from {version_paths}",
+                data={"raw_output": res.stdout},
+                priority=EventPriority.ERROR,
             )
-            self.result.message = f"ROCm: {rocm_data.model_dump()}"
-            self.result.status = ExecutionStatus.OK
-        else:
-            rocm_data = None
+
+        if not rocm_data:
             self._log_event(
                 category=EventCategory.OS,
                 description="Error checking ROCm version",

nodescraper/plugins/inband/rocm/rocmdata.py

@@ -48,12 +48,6 @@ def validate_rocm_version(cls, rocm_version: str) -> str:
         Returns:
             str: The validated ROCm version string.
         """
-        if not bool(
-            re.match(
-                r"^(\d+(?:\.\d+){0,3})-(\d+)$",
-                rocm_version,
-            )
-        ):
+        if not re.match(r"^\d+(?:\.\d+){0,3}(-\d+)?$", rocm_version):
             raise ValueError(f"ROCm version has invalid format: {rocm_version}")
-
         return rocm_version