From 1f793e8750c84ac6f7c509415bc5adaa4b37ff92 Mon Sep 17 00:00:00 2001
From: Nawaz Dhandala <hello@nawazdhandala.com>
Date: Mon, 29 Dec 2025 15:54:18 +0000
Subject: [PATCH] feat: Add job to reset stuck AI Agent tasks to Scheduled
 status

---
 Worker/Jobs/AIAgent/TimeoutStuckTasks.ts | 81 ++++++++++++++++++++++++
 Worker/Routes.ts                         |  1 +
 2 files changed, 82 insertions(+)
 create mode 100644 Worker/Jobs/AIAgent/TimeoutStuckTasks.ts

diff --git a/Worker/Jobs/AIAgent/TimeoutStuckTasks.ts b/Worker/Jobs/AIAgent/TimeoutStuckTasks.ts
new file mode 100644
index 0000000000..7d40a3eec6
--- /dev/null
+++ b/Worker/Jobs/AIAgent/TimeoutStuckTasks.ts
@@ -0,0 +1,81 @@
+import RunCron from "../../Utils/Cron";
+import OneUptimeDate from "Common/Types/Date";
+import AIAgentTaskStatus from "Common/Types/AI/AIAgentTaskStatus";
+import { EVERY_MINUTE } from "Common/Utils/CronTime";
+import AIAgentTaskService from "Common/Server/Services/AIAgentTaskService";
+import QueryHelper from "Common/Server/Types/Database/QueryHelper";
+import AIAgentTask from "Common/Models/DatabaseModels/AIAgentTask";
+import logger from "Common/Server/Utils/Logger";
+
+/**
+ * AI Agent tasks that are stuck in "InProgress" status for more than 30 minutes
+ * are considered timed out. This can happen when the AI Agent container crashes
+ * or is deleted while processing a task.
+ *
+ * This job resets stuck tasks back to "Scheduled" status so they can be retried.
+ */
+
+const TASK_TIMEOUT_MINUTES: number = 30;
+
+RunCron(
+  "AIAgent:TimeoutStuckTasks",
+  {
+    schedule: EVERY_MINUTE,
+    runOnStartup: false,
+  },
+  async () => {
+    logger.debug("Checking for stuck AI Agent tasks");
+
+    const timeoutThreshold: Date =
+      OneUptimeDate.getSomeMinutesAgo(TASK_TIMEOUT_MINUTES);
+
+    // Find tasks that have been InProgress for longer than the timeout threshold
+    const stuckTasks: Array<AIAgentTask> = await AIAgentTaskService.findAllBy({
+      query: {
+        status: AIAgentTaskStatus.InProgress,
+        startedAt: QueryHelper.lessThan(timeoutThreshold),
+      },
+      select: {
+        _id: true,
+        startedAt: true,
+      },
+      props: {
+        isRoot: true,
+      },
+    });
+
+    if (stuckTasks.length === 0) {
+      return;
+    }
+
+    logger.info(
+      `Found ${stuckTasks.length} stuck AI Agent task(s). Resetting to Scheduled status.`,
+    );
+
+    for (const task of stuckTasks) {
+      try {
+        await AIAgentTaskService.updateOneById({
+          id: task.id!,
+          data: {
+            status: AIAgentTaskStatus.Scheduled,
+            statusMessage:
+              "Task was reset due to timeout. The AI Agent processing this task may have crashed or been terminated.",
+            startedAt: null,
+          },
+          props: {
+            isRoot: true,
+          },
+        });
+
+        logger.info(
+          `Reset stuck AI Agent task ${task.id?.toString()} to Scheduled status`,
+        );
+      } catch (error) {
+        logger.error(
+          `Failed to reset stuck AI Agent task ${task.id?.toString()}:`,
+        );
+        logger.error(error);
+      }
+    }
+  },
+);
diff --git a/Worker/Routes.ts b/Worker/Routes.ts
index 60924e13bd..7ac5e159c2 100644
--- a/Worker/Routes.ts
+++ b/Worker/Routes.ts
@@ -90,6 +90,7 @@ import "./Jobs/Probe/UpdateConnectionStatus";
 // AI Agents
 import "./Jobs/AIAgent/SendOwnerAddedNotification";
 import "./Jobs/AIAgent/UpdateConnectionStatus";
+import "./Jobs/AIAgent/TimeoutStuckTasks";
 
 // Telemetry Monitors.
 import "./Jobs/TelemetryMonitor/MonitorTelemetryMonitor";