ci(ariadne): emit per-test case result metrics for flaky tracking

scripts/publish_test_metrics.py

@@ -65,6 +65,37 @@ def _load_junit(path: str) -> dict[str, int]:
     return totals
 
 
+def _load_junit_cases(path: str) -> list[tuple[str, str]]:
+    tree = ET.parse(path)
+    root = tree.getroot()
+
+    suites: list[ET.Element]
+    if root.tag == "testsuite":
+        suites = [root]
+    elif root.tag == "testsuites":
+        suites = list(root.findall("testsuite"))
+    else:
+        suites = []
+
+    cases: list[tuple[str, str]] = []
+    for suite in suites:
+        for case in suite.findall("testcase"):
+            name = (case.attrib.get("name") or "").strip()
+            classname = (case.attrib.get("classname") or "").strip()
+            if not name:
+                continue
+            test_id = f"{classname}::{name}" if classname else name
+            status = "passed"
+            if case.find("failure") is not None:
+                status = "failed"
+            elif case.find("error") is not None:
+                status = "error"
+            elif case.find("skipped") is not None:
+                status = "skipped"
+            cases.append((test_id, status))
+    return cases
+
+
 def _read_http(url: str) -> str:
     try:
         with urllib.request.urlopen(url, timeout=10) as resp:
@@ -192,8 +223,10 @@ def main() -> int:
     docs_gate_rc = _load_gate_rc(Path(os.getenv("QUALITY_GATE_DOCS_RC_PATH", str(build_dir / "docs-naming.rc"))))
     source_lines_over_500 = _count_source_files_over_limit(repo_root, max_lines=500)
     totals = {"tests": 0, "failures": 0, "errors": 0, "skipped": 0}
+    test_cases: list[tuple[str, str]] = []
     if os.path.exists(junit_path):
         totals = _load_junit(junit_path)
+        test_cases = _load_junit_cases(junit_path)
     passed = max(totals["tests"] - totals["failures"] - totals["errors"] - totals["skipped"], 0)
 
     outcome = "ok"
@@ -247,9 +280,14 @@ def main() -> int:
         "# TYPE platform_quality_gate_source_lines_over_500_total gauge",
         f'platform_quality_gate_source_lines_over_500_total{{suite="{suite}"}} {source_lines_over_500}',
         "# TYPE ariadne_quality_gate_checks_total gauge",
+        "# TYPE platform_quality_gate_test_case_result gauge",
         "# TYPE ariadne_quality_gate_build_info gauge",
         f"ariadne_quality_gate_build_info{_label_str(labels)} 1",
     ]
+    payload_lines.extend(
+        f'platform_quality_gate_test_case_result{{suite="{suite}",test="{_escape_label(test_name)}",status="{_escape_label(test_status)}"}} 1'
+        for test_name, test_status in test_cases
+    )
     payload_lines.extend(
         f'ariadne_quality_gate_checks_total{{suite="{suite}",check="{check_name}",result="{check_status}"}} 1'
         for check_name, check_status in checks.items()