ci(pegasus): emit test-case status metrics for flaky-test tracking

scripts/publish_test_metrics.py

@@ -8,6 +8,7 @@ Inputs:
 Outputs pushed:
 - platform_quality_gate_runs_total{suite="pegasus",status="ok|failed"}
 - pegasus_quality_gate_tests_total{suite="pegasus",result=*}
+- platform_quality_gate_test_case_result{suite="pegasus",test=*,status=*}
 - pegasus_quality_gate_coverage_percent{suite="pegasus"}
 - platform_quality_gate_workspace_line_coverage_percent{suite="pegasus"}
 - platform_quality_gate_source_lines_over_500_total{suite="pegasus"}
@@ -88,18 +89,20 @@ def _load_junit_cases(path: Path) -> list[tuple[str, str]]:
 
     cases: list[tuple[str, str]] = []
     for suite in suites:
-        for test_case in suite.findall("testcase"):
+        for case in suite.findall("testcase"):
-            case_name = (test_case.attrib.get("name") or "").strip()
+            name = (case.attrib.get("name") or "").strip()
-            class_name = (test_case.attrib.get("classname") or "").strip()
+            classname = (case.attrib.get("classname") or "").strip()
-            if not case_name:
+            if not name:
                 continue
-            full_name = f"{class_name}.{case_name}" if class_name else case_name
+            test_id = f"{classname}::{name}" if classname else name
             status = "passed"
-            if test_case.find("failure") is not None or test_case.find("error") is not None:
+            if case.find("failure") is not None:
                 status = "failed"
-            elif test_case.find("skipped") is not None:
+            elif case.find("error") is not None:
+                status = "error"
+            elif case.find("skipped") is not None:
                 status = "skipped"
-            cases.append((full_name, status))
+            cases.append((test_id, status))
     return cases
 
 
@@ -259,9 +262,7 @@ def main() -> int:
 
     b = _load_junit(backend_junit)
     f = _load_junit(frontend_junit)
-    backend_cases = _load_junit_cases(backend_junit)
+    test_cases = _load_junit_cases(backend_junit) + _load_junit_cases(frontend_junit)
-    frontend_cases = _load_junit_cases(frontend_junit)
-    test_cases = backend_cases + frontend_cases
     totals = {
         "tests": b["tests"] + f["tests"],
         "failures": b["failures"] + f["failures"],
@@ -351,6 +352,7 @@ def main() -> int:
         "# TYPE pegasus_quality_gate_issues_total gauge",
         f'pegasus_quality_gate_issues_total{{suite="{suite}"}} {len(gate_issues)}',
         "# TYPE pegasus_quality_gate_checks_total gauge",
+        "# TYPE platform_quality_gate_test_case_result gauge",
         "# TYPE pegasus_quality_gate_build_info gauge",
         f"pegasus_quality_gate_build_info{_label_str(labels)} 1",
     ]
@@ -358,7 +360,6 @@ def main() -> int:
         f'pegasus_quality_gate_checks_total{{suite="{suite}",check="{check_name}",result="{check_status}"}} 1'
         for check_name, check_status in checks.items()
     )
-    payload_lines.append("# TYPE platform_quality_gate_test_case_result gauge")
     if test_cases:
         payload_lines.extend(
             [