fix(atvm-watcher): synthesize failed host result on hang-kill/nonzero exit; update run learning and vm inventory

atvm/watcher-service/atvm_run_watcher.py

@@ -1212,6 +1212,30 @@ def infer_host_from_subrun_build(
     return remaining_hosts[0] if remaining_hosts else None
 
 
+def synthesize_failed_host_result(
+    build_name: str,
+    log_text: str,
+    inventory: Dict[str, str],
+    reason: str,
+) -> Dict[str, HostResult]:
+    expected_hosts = extract_expected_hosts(log_text)
+    host = find_current_running_host(log_text, [])
+    if not host and expected_hosts:
+        host = expected_hosts[0]
+    if not host:
+        host = build_name
+    return {
+        host: HostResult(
+            host=host,
+            kernel=inventory.get(host, "unknown"),
+            status="FAIL",
+            detail=reason,
+            tests=1,
+            failures=1,
+        )
+    }
+
+
 def infer_metadata(build_name: str, log_text: str) -> Dict[str, object]:
     try:
         extra_options = json.loads(os.environ.get("ATVM_WATCHER_EXTRA_OPTIONS", "[]"))
@@ -2131,6 +2155,15 @@ def determine_state(
         return "RUNNING", subrun_states, parent_host_results, start_ts, end_ts, currents_url, notes
 
     if hang_kill_detected:
+        if not parent_host_results:
+            parent_host_results = synthesize_failed_host_result(
+                build_name=build_name,
+                log_text=log_text,
+                inventory=inventory,
+                reason="hang timeout killed runner",
+            )
+            if subrun_states:
+                subrun_states[-1]["host_results"] = dict(parent_host_results)
         notes.append("Runner hang monitor terminated the run (`SIGKILL`) after inactivity.")
         notes.append("Treating this as explicit failure.")
         return "FAILED", subrun_states, parent_host_results, start_ts, end_ts, currents_url, notes
@@ -2146,6 +2179,15 @@ def determine_state(
 
     if process_gone_since and (now_utc() - process_gone_since).total_seconds() >= process_exit_grace_seconds:
         if runner_result == "exit-code" or (runner_exit_code is not None and runner_exit_code != 0):
+            if not parent_host_results:
+                parent_host_results = synthesize_failed_host_result(
+                    build_name=build_name,
+                    log_text=log_text,
+                    inventory=inventory,
+                    reason="runner exited non-zero before host results were emitted",
+                )
+                if subrun_states:
+                    subrun_states[-1]["host_results"] = dict(parent_host_results)
             detail = f"Runner service ended with non-zero exit ({runner_exit_code})"
             if runner_active_state:
                 detail += f"; state={runner_active_state}"