ci: stabilize dind and publish quality metrics via pushgateway

Jenkinsfile

@@ -61,8 +61,7 @@ spec:
     - name: docker-config-writable
       emptyDir: {}
     - name: dind-storage
-      persistentVolumeClaim:
+      emptyDir: {}
-        claimName: jenkins-dind-cache
     - name: harbor-config
       secret:
         secretName: harbor-robot-pipeline
@@ -80,9 +79,6 @@ spec:
     COVERAGE_MIN = '99'
     COVERAGE_JSON = 'build/coverage.json'
     JUNIT_XML = 'build/junit.xml'
-    METRICS_PREFIX = 'ariadne_ci'
-    VM_IMPORT_URL = 'http://victoria-metrics-single-server.monitoring.svc.cluster.local:8428/api/v1/import/prometheus'
-    REPO_NAME = 'ariadne'
     SUITE_NAME = 'ariadne'
     PUSHGATEWAY_URL = 'http://platform-quality-gateway.monitoring.svc.cluster.local:9091'
   }
@@ -165,13 +161,23 @@ python -c "import json; payload=json.load(open('build/coverage.json', encoding='
         container('builder') {
           sh '''
             set -euo pipefail
+            ready=0
             for i in $(seq 1 10); do
               if docker info >/dev/null 2>&1; then
+                ready=1
                 break
               fi
               sleep 2
             done
-            docker buildx use default || docker buildx create --name default --driver docker --use
+            if [ "${ready}" -ne 1 ]; then
+              echo "docker daemon did not become ready on ${DOCKER_HOST}" >&2
+              docker version || true
+              exit 1
+            fi
+            BUILDER_NAME="ariadne-${BUILD_NUMBER}"
+            docker buildx rm "${BUILDER_NAME}" >/dev/null 2>&1 || true
+            docker buildx create --name "${BUILDER_NAME}" --driver docker-container --bootstrap --use
+            docker buildx inspect "${BUILDER_NAME}" --bootstrap
           '''
         }
       }

scripts/publish_test_metrics.py

@@ -1,4 +1,6 @@
 #!/usr/bin/env python3
+"""Publish Ariadne quality-gate test metrics to Pushgateway (Prometheus ingest)."""
+
 from __future__ import annotations
 
 import json
@@ -31,18 +33,20 @@ def _load_junit(path: str) -> dict[str, int]:
     tree = ET.parse(path)
     root = tree.getroot()
 
-    def _as_int(node, name: str) -> int:
+    def _as_int(node: ET.Element, name: str) -> int:
         raw = node.attrib.get(name) or "0"
         try:
             return int(float(raw))
         except ValueError:
             return 0
 
-    suites = []
+    suites: list[ET.Element]
     if root.tag == "testsuite":
         suites = [root]
     elif root.tag == "testsuites":
         suites = list(root.findall("testsuite"))
+    else:
+        suites = []
 
     totals = {"tests": 0, "failures": 0, "errors": 0, "skipped": 0}
     for suite in suites:
@@ -53,7 +57,15 @@ def _load_junit(path: str) -> dict[str, int]:
     return totals
 
 
-def _post_metrics(url: str, payload: str) -> None:
+def _read_http(url: str) -> str:
+    try:
+        with urllib.request.urlopen(url, timeout=10) as resp:
+            return resp.read().decode("utf-8", errors="replace")
+    except Exception:
+        return ""
+
+
+def _post_text(url: str, payload: str) -> None:
     req = urllib.request.Request(
         url,
         data=payload.encode("utf-8"),
@@ -65,14 +77,36 @@ def _post_metrics(url: str, payload: str) -> None:
             raise RuntimeError(f"metrics push failed status={resp.status}")
 
 
-def main() -> int:
+def _fetch_existing_counter(pushgateway_url: str, metric: str, labels: dict[str, str]) -> float:
-    vm_url = os.getenv("VM_IMPORT_URL", "").strip()
+    text = _read_http(f"{pushgateway_url.rstrip('/')}/metrics")
-    if not vm_url:
+    if not text:
-        print("VM_IMPORT_URL not set; skipping metrics push")
+        return 0.0
-        return 0
 
+    for line in text.splitlines():
+        if not line.startswith(metric + "{"):
+            continue
+        if any(f'{k}="{v}"' not in line for k, v in labels.items()):
+            continue
+        parts = line.split()
+        if len(parts) < 2:
+            continue
+        try:
+            return float(parts[1])
+        except ValueError:
+            return 0.0
+    return 0.0
+
+
+def main() -> int:
     coverage_path = os.getenv("COVERAGE_JSON", "build/coverage.json")
     junit_path = os.getenv("JUNIT_XML", "build/junit.xml")
+    pushgateway_url = os.getenv(
+        "PUSHGATEWAY_URL", "http://platform-quality-gateway.monitoring.svc.cluster.local:9091"
+    ).strip()
+    suite = os.getenv("SUITE_NAME", "ariadne")
+    branch = os.getenv("BRANCH_NAME", "")
+    build_number = os.getenv("BUILD_NUMBER", "")
+    commit = os.getenv("GIT_COMMIT", "")
 
     if not os.path.exists(coverage_path):
         raise RuntimeError(f"missing coverage file {coverage_path}")
@@ -83,33 +117,72 @@ def main() -> int:
     totals = _load_junit(junit_path)
     passed = max(totals["tests"] - totals["failures"] - totals["errors"] - totals["skipped"], 0)
 
+    outcome = "ok"
+    if totals["tests"] <= 0 or totals["failures"] > 0 or totals["errors"] > 0:
+        outcome = "failed"
+
+    job_name = "platform-quality-ci"
+    ok_count = _fetch_existing_counter(
+        pushgateway_url,
+        "platform_quality_gate_runs_total",
+        {"job": job_name, "suite": suite, "status": "ok"},
+    )
+    failed_count = _fetch_existing_counter(
+        pushgateway_url,
+        "platform_quality_gate_runs_total",
+        {"job": job_name, "suite": suite, "status": "failed"},
+    )
+    if outcome == "ok":
+        ok_count += 1
+    else:
+        failed_count += 1
+
     labels = {
-        "job": os.getenv("CI_JOB_NAME", "ariadne"),
+        "suite": suite,
-        "branch": os.getenv("BRANCH_NAME", ""),
+        "branch": branch,
-        "build_number": os.getenv("BUILD_NUMBER", ""),
+        "build_number": build_number,
-        "commit": os.getenv("GIT_COMMIT", ""),
+        "commit": commit,
-        "repo": os.getenv("REPO_NAME", "ariadne"),
     }
-
+    payload_lines = [
-    prefix = os.getenv("METRICS_PREFIX", "ariadne_ci")
+        "# TYPE platform_quality_gate_runs_total counter",
-    lines = [
+        f'platform_quality_gate_runs_total{{suite="{suite}",status="ok"}} {ok_count:.0f}',
-        f"{prefix}_coverage_percent{_label_str(labels)} {coverage:.3f}",
+        f'platform_quality_gate_runs_total{{suite="{suite}",status="failed"}} {failed_count:.0f}',
-        f"{prefix}_tests_total{_label_str({**labels, 'result': 'passed'})} {passed}",
+        "# TYPE ariadne_quality_gate_tests_total gauge",
-        f"{prefix}_tests_total{_label_str({**labels, 'result': 'failed'})} {totals['failures']}",
+        f'ariadne_quality_gate_tests_total{{suite="{suite}",result="passed"}} {passed}',
-        f"{prefix}_tests_total{_label_str({**labels, 'result': 'error'})} {totals['errors']}",
+        f'ariadne_quality_gate_tests_total{{suite="{suite}",result="failed"}} {totals["failures"]}',
-        f"{prefix}_tests_total{_label_str({**labels, 'result': 'skipped'})} {totals['skipped']}",
+        f'ariadne_quality_gate_tests_total{{suite="{suite}",result="error"}} {totals["errors"]}',
-        f"{prefix}_build_info{_label_str(labels)} 1",
+        f'ariadne_quality_gate_tests_total{{suite="{suite}",result="skipped"}} {totals["skipped"]}',
+        "# TYPE ariadne_quality_gate_coverage_percent gauge",
+        f'ariadne_quality_gate_coverage_percent{{suite="{suite}"}} {coverage:.3f}',
+        "# TYPE ariadne_quality_gate_build_info gauge",
+        f"ariadne_quality_gate_build_info{_label_str(labels)} 1",
     ]
+    payload = "\n".join(payload_lines) + "\n"
+    _post_text(f"{pushgateway_url.rstrip('/')}/metrics/job/{job_name}/suite/{suite}", payload)
 
-    payload = "\n".join(lines) + "\n"
+    print(
-    _post_metrics(vm_url, payload)
+        json.dumps(
-    print("metrics push complete")
+            {
+                "suite": suite,
+                "outcome": outcome,
+                "tests_total": totals["tests"],
+                "tests_passed": passed,
+                "tests_failed": totals["failures"],
+                "tests_errors": totals["errors"],
+                "tests_skipped": totals["skipped"],
+                "coverage_percent": round(coverage, 3),
+                "ok_counter": ok_count,
+                "failed_counter": failed_count,
+            },
+            indent=2,
+        )
+    )
     return 0
 
 
 if __name__ == "__main__":
     try:
-        sys.exit(main())
+        raise SystemExit(main())
     except Exception as exc:
         print(f"metrics push failed: {exc}")
-        sys.exit(1)
+        raise