When you retire unused tasks or tables in DataWorks, direct undeployment gives you no chance to assess downstream impact, notify affected teams, or reverse a mistake. Graceful undeployment adds a managed, multi-stage process: before any object is removed, the system runs an impact assessment so you can see which downstream dependencies are at risk. You can then configure notifications, choose an execution mode, and roll back at any point before the actual undeployment takes effect.
Use graceful undeployment instead of direct undeployment when the task or table has downstream consumers, when you need a review window before committing, or when you want an audit trail across teams.
How it works
A graceful undeployment plan moves through up to eight sequential stages. You select which optional stages to include; mandatory stages run automatically.
| Stage | Default | Can skip | What happens |
|---|---|---|---|
| Start undeployment | Yes | No | Starts the plan. |
| Delayed scheduling | No | Yes | Extends the scheduled run time of the target object, delaying data generation for it and its downstream dependencies. |
| Suspend scheduling | No | Yes | Freezes the object. Its status becomes frozen in Operation Center > Auto Triggered Nodes, and it stops generating table data along with its downstream objects. |
| Undeploy task | Yes | Yes | Undeploys the object. No new instances are generated starting the next day; existing instances are set to zero load. |
| Rename output table | No | Yes | Renames the output table to mark it as belonging to an undeployed object. |
| Reset table lifecycle | No | Yes | Resets the lifecycle of the output table. MaxCompute tables only. |
| Delete output table | No | Yes | Deletes the output table after undeployment. |
| Undeploy completed | Yes | No | Ends the plan. |
Rollback boundary: For task objects, rollback is available until the Undeploy task step completes. For table objects, rollback is available until the Undeploy completed step. Individual step rollback is not supported — rollback reverts all completed steps for that object.
Rollback is not available if the plan has been canceled, a substep is in progress, or the Undeploy task step has already completed.
Limitations
An undeployment plan can only include objects from a single workspace.
A plan can contain a maximum of 200 objects.
Objects already included in an active undeployment plan cannot be added to another plan.
The following objects cannot be selected for undeployment: the root node of a workspace, and virtual nodes created by an undeployment plan.
If a task being undeployed has downstream dependencies whose scheduling dependencies are not removed in time, the system creates a virtual node and attaches those downstream tasks to it.
The supported stages for each node type may vary. Refer to the product UI for details.
Prerequisites
Before you begin, make sure you have one of the following roles in the target workspace:
| Role | Create plan | Configure plan | Start / execute plan | Full operations |
|---|---|---|---|---|
| Alibaba Cloud account | Yes | Yes | Yes | Yes |
RAM user (AliyunDataWorksFullAccess) | Yes | Yes | Yes | Yes |
| Workspace administrator | Yes | Yes | Yes | Yes |
| Tenant-level data governance administrator | Yes | Yes | No — cannot start or execute plans containing tasks for which they lack permissions | No — can create, configure, and delete plans only |
| Workspace-level data governance administrator | Yes | Yes | No — cannot start or execute plans containing tasks for which they lack permissions | No — can create, configure, and delete plans only |
| O&M engineer | Yes — own plans only | Yes — own plans only | Yes — own plans only | No — read-only for others' plans |
Create an undeployment plan
Step 1: Open the Shutdown page
Log on to the DataWorks console. In the top navigation bar, select the target region.
In the left-side navigation pane, choose Data Governance > Data Asset Governance, then click Go to Data Asset Governance.
In the left-side navigation pane, choose Governance > Automation > Shutdown.
From the Workspace drop-down list at the top of the page, select the workspace that contains the objects you want to undeploy. The list shows only workspaces you have joined. If you need to join a workspace, ask the workspace administrator to add you. See Add workspace members and manage their roles and permissions.
Click Create Plan. Alternatively, on the Process Governance Items page, find the task or table you want to undeploy and click Create Plan in the Actions column.
Step 2: Select objects to undeploy
In the Plan Name field, enter a name for the plan. Confirm the Workspace field shows the correct workspace.
In the To-Be-Undeployed Object section, add the objects you want to undeploy. Objects can be published tasks or tables. For tasks, use either method: For tables, use Add To-Be-Undeployed Object.
Add To-Be-Undeployed Object — add a task by its node ID. Find the node ID on the task configuration page under Scheduling > Basic Properties, or on the Auto Triggered Node O&M > Auto Triggered Nodes page in the Operation Center.
Import Based on Lineage — select a task, then select related tasks based on their data lineage hierarchy.
Review the impact assessment for each added object:
Field Description Undeployment impact assessment Low — the output tables have not been consumed in the last month. High — the output tables have been consumed in the last month. Impact item description A plain-language explanation of the impact level. The output table is consumed The consumption status of the output table for the task object. Task that generates table The task that produces the table object. Click to view task details. Number of affected baselines/highest baseline level The number of affected baselines and the highest baseline level, for both the task and the table object. See Baseline management.
Step 3: Configure the process
Select which stages to include and configure notifications for each stage.
The plan always starts with Start undeployment and ends with Undeploy completed. Toggle optional stages on or off based on your scenario — any deselected stage is skipped.
At each stage, you can send notifications to specified users through:
| Channel | Notes |
|---|---|
| System Prompt | In-console message. View at Data Asset Governance > Overview > Reminder > Governance Reminder. |
| Sent to the specified recipients. | |
| DingTalk Chatbot | Requires a configured DingTalk chatbot. See Scenario practice: Send alert notifications to a DingTalk group. |
| Webhook | Sends a POST request to the configured endpoint. |
Step 4: Set the execution mode
Choose how each stage advances to the next:
| Mode | How it works | When to use |
|---|---|---|
| Manually trigger execution | Each stage waits for manual confirmation from the plan creator before proceeding. | Tasks or tables critical to your business workflow. |
| Automatic execution | Each stage proceeds automatically after a configured waiting period. | Batch undeployment of low-priority workflows. |
Step 5: Manage the plan
On the Shutdown page, you can view all undeployment plans and take the following actions on a plan:
Execute — starts the plan.
Cancel — stops any stages currently in progress. Stages that have already completed are not affected.
Delete — cancels any in-progress stages, then deletes the plan. This action cannot be undone.
To inspect or act on individual objects within a plan, click Actions > Details:
| Action | Description |
|---|---|
| View execution details | See the progress, status, and details of each stage for a single object. |
| Cancel task | Cancel any incomplete stages for that object. |
| Rollback task | Revert all completed stages for that object. Not available if the plan is canceled, a substep is in progress, or the Undeploy task step has already completed. |
| Go to O&M | Open the object in Operation Center. |
| Edit | Open the task in DataStudio for editing. |