AI Diagnose for tasks uses Buddy AI Assistant to automatically analyze failure causes, impact scope, and repair suggestions for tasks in Failed, Waiting for Resources, or Upstream Failed states, helping users quickly locate issues.
Overview
Scope of AI Diagnose:
|
Failed | ✅ | Error Classification, Root Cause, and Impact Scope |
Waiting for Resources | ✅ | Resource queuing / Insufficient resources / Quota exceeded / Concurrency limit reached / Abnormal resource group configuration / Oversized task request |
Upstream Failed | ✅ | Failed Link Tracing, Root Cause Task Identification, and Impact Scope |
Other statuses (succeeded, running, skipped, and so on) | ❌ | - |
Prerequisites
With View or higher permission on the task, diagnosis can be triggered.
Buddy has been enabled for the workspace. For details, see Buddy. Operation Steps
Step 1: Find the AI Diagnosis Entry
AI Diagnose entries are available in multiple locations. Select an appropriate entry based on the view you are currently on:
|
Workflow Run Details - DAG Graph | hover over a node in the failed / waiting for resources / upstream failed state, and click AI Diagnosis in the pop-up. |
Workflow Run Details - List Mode | AI Diagnosis after the failed / waiting for resources / upstream failed status column in the list row |
Workflow Run Details - Gantt Chart | hover over a node in the failed / waiting for resources / upstream failed state in the Gantt chart, and click AI Diagnosis in the pop-up. |
Workflow Details Page List - Run History Graph | hover over a node in the failed / waiting for resources / upstream failed state, and click AI Diagnosis in the pop-up. |
Task Run Details Page | AI Diagnosis button in the top operation bar |
Step 2: Trigger Diagnosis
Click AI Diagnose to automatically open Buddy Assistant and start diagnosis.
Note:
When a task has multiple runs:
From the workflow list / DAG / list, or Gantt chart entry, trigger diagnosis of the last task run.
From the task run details page entry, trigger diagnosis of the currently selected task run.
Step 3: View Diagnosis Results
After diagnosis is complete, Buddy displays the following:
Area | Content. |
Instance information | Task name, task run ID, current status, and more |
Issue summary | Problem category |
Root cause analysis | Root cause conclusion + detailed analysis process |
Impact Scope | Number and list of direct downstream tasks (expandable) |
Fixing recommendations | Fix solution provided by AI |
Step 4: Diagnosis Feedback
On the right side of the diagnosis result, you can:
Operation | Description |
Helpful | Mark the diagnosis as helpful, and the feedback data is used for model supervised learning. |
Inaccurate | Mark the diagnosis as inaccurate. |
Rediagnose | Run the diagnosis again. |
Feedback data helps Buddy continuously improve diagnosis results.
Description
Permission | Trigger diagnosis | Perform repair operations |
View. | ✅ | ❌(View recommendations only) |
Run / Manage | ✅ | ✅ |
If the suggestion involves modifying the task configuration or rerunning the task, and the current user has only View permission, contact the administrator for assistance.
Use Limits
AI Diagnose takes effect only for the following three states: Failed, Waiting for Resources, and Upstream Failed.
The diagnosis result is generated by AI and is for reference only. Manual review is recommended.
If the logs of previously failed tasks have been cleaned up (beyond the retention period), the diagnosis information will be incomplete.
FAQs
Q1: Why does the diagnosis suggest "Resource Queuing" when resources are actually sufficient?
The resource group load may have decreased after the diagnosis was completed. Recommendations:
Click Diagnose Again to obtain the latest status.
Q2: Can the root cause task be directly rerun during upstream failure diagnosis?
You can initiate a rerun in Buddy based on the diagnosis result. The subsequent behavior of the rerun is the same as that of a manual rerun. For details, see Workflow and Task Ops. Q3: Which languages does AI Diagnose support for international sites and overseas users?
Buddy supports both Chinese and English input and output. Technical fields such as task names and error codes remain unchanged for easier troubleshooting.
Q4: What should I do if no diagnosis result is returned?
Check whether Buddy loads properly. You may need to log in again.
Check whether the task logs are complete: if logs are truncated or retained beyond the expiration period, the diagnosis accuracy will decrease.
You can click Diagnose Again to retry. If failures persist, contact the workspace administrator.
References