atlasbot: expose job failure metrics

atlasbot/snapshot/builder.py

@@ -660,6 +660,23 @@ def _append_restarts(lines: list[str], summary: dict[str, Any]) -> None:
         lines.append("restarts_1h_top: " + "; ".join(parts))
 
 
+def _append_job_failures(lines: list[str], summary: dict[str, Any]) -> None:
+    metrics = summary.get("metrics") if isinstance(summary.get("metrics"), dict) else {}
+    failures = metrics.get("job_failures_24h") if isinstance(metrics.get("job_failures_24h"), list) else []
+    if not failures:
+        return
+    parts = []
+    for entry in failures:
+        metric = entry.get("metric") if isinstance(entry, dict) else {}
+        namespace = metric.get("namespace")
+        job_name = metric.get("job_name") or metric.get("job")
+        value = entry.get("value")
+        if namespace and job_name and value is not None:
+            parts.append(f"{namespace}/{job_name}={_format_float(value)}")
+    if parts:
+        lines.append("job_failures_24h: " + "; ".join(parts))
+
+
 def _append_postgres(lines: list[str], summary: dict[str, Any]) -> None:
     postgres = summary.get("postgres") if isinstance(summary.get("postgres"), dict) else {}
     if not postgres:
@@ -762,6 +779,7 @@ def summary_text(snapshot: dict[str, Any] | None) -> str:
     _append_namespace_usage(lines, summary)
     _append_pod_usage(lines, summary)
     _append_restarts(lines, summary)
+    _append_job_failures(lines, summary)
     _append_postgres(lines, summary)
     _append_hottest(lines, summary)
     _append_pvc_usage(lines, summary)