unit test update

Author: landrews-amd

nodescraper/plugins/inband/dmesg/dmesg_analyzer.py

@@ -413,18 +413,25 @@ def _is_known_error(self, known_err_events: list[Event], unknown_match: str) ->
             bool: return True if error is known
         """
         for regex_obj in self.ERROR_REGEX:
-            if regex_obj.regex.search(unknown_match):
-                return True
+            try:
+                if regex_obj.regex.search(unknown_match):
+                    return True
+            except re.error:
+                continue
 
-        # handle multline matches
         for event in known_err_events:
             known_match = event.data["match_content"]
             if isinstance(known_match, list):
                 for line in known_match:
-                    if unknown_match in line:
+                    if unknown_match == line or unknown_match in line or line in unknown_match:
                         return True
-            elif known_match in unknown_match:
-                return True
+            elif isinstance(known_match, str):
+                if (
+                    unknown_match == known_match
+                    or unknown_match in known_match
+                    or known_match in unknown_match
+                ):
+                    return True
         return False
 
     def analyze_data(

test/unit/plugin/test_dmesg_analyzer.py

@@ -234,3 +234,48 @@ def test_lnet_and_lustre_boot_errors_are_warning_events(system_info):
     for m in (m1, m2, m3):
         ev = by_msg[m]
         assert ev.priority == EventPriority.WARNING, f"{m} should be WARNING"
+
+
+def test_aca(system_info):
+    aca_data1 = DmesgData(
+        dmesg_content=(
+            "kern  :err   : 2025-01-01T10:17:15,145363-04:00 amdgpu 0000:0c:00.0: amdgpu: [Hardware error] Accelerator Check Architecture events logged\n"
+            "kern  :err   : 2025-01-01T10:17:15,145363-04:00 amdgpu 0000:0c:00.0: amdgpu: [Hardware error] aca entry[00].STATUS=0x000000000000000f\n"
+            "kern  :err   : 2025-01-01T10:17:15,145363-04:00 amdgpu 0000:0c:00.0: amdgpu: [Hardware error] aca entry[00].ADDR=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T10:17:15,145363-04:00 amdgpu 0000:0c:00.0: amdgpu: [Hardware error] aca entry[00].MISC0=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T10:17:15,145363-04:00 amdgpu 0000:0c:00.0: amdgpu: [Hardware error] aca entry[00].IPID=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T10:17:15,145363-04:00 amdgpu 0000:0c:00.0: amdgpu: [Hardware error] aca entry[00].SYND=0x0000000000000000\n"
+        )
+    )
+
+    aca_data2 = DmesgData(
+        dmesg_content=(
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: Accelerator Check Architecture events logged\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].CONTROL=0x000000000000000f\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].STATUS=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].ADDR=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].MISC=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].CONFIG=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].IPID=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].SYND=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].DESTAT=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].DEADDR=0x0000000000000000\n"
+            "kern  :err   : 2025-01-01T17:53:23,028841-06:00 amdgpu 0000:48:00.0: {1}[Hardware Error]: ACA[01/01].CONTROL_MASK=0x0000000000000000\n"
+        )
+    )
+
+    analyzer = DmesgAnalyzer(
+        system_info=system_info,
+    )
+
+    res = analyzer.analyze_data(aca_data1)
+    assert res.status == ExecutionStatus.ERROR
+    assert len(res.events) == 1
+    assert res.events[0].description == "ACA Error"
+    assert res.events[0].priority == EventPriority.ERROR
+
+    res = analyzer.analyze_data(aca_data2)
+    assert res.status == ExecutionStatus.ERROR
+    assert len(res.events) == 1
+    assert res.events[0].description == "ACA Error"
+    assert res.events[0].priority == EventPriority.ERROR