The DAG (Directed Acyclic Graph) visualizes the dependency relationships between scheduled nodes. To open a node's DAG, click DAG Graph next to the node in Operation Center.
The DAG opens in a three-layer view by default: the current node, its first-level parent nodes, and its first-level child nodes. This topic uses the DAG of a recurring instance as the reference example.
Limits
Node aggregation, Upstream Analysis, and Downstream Analysis require DataWorks Professional Edition or higher. For upgrade options, see Software edition selection and billing.
How it works
A solid line between nodes indicates a dependency within the same scheduling epoch. A dashed line indicates a cross-cycle dependency that spans scheduling epochs.
A node's state determines which operations you can perform on it and whether it blocks downstream nodes from running. Use the following table to quickly determine what actions are available for a stuck or failed node.
State | Blocks downstream? | Available operations |
Running | Yes | Stop, Pause (Freeze), Emergency Operation |
Failed | Yes | Rerun, Rerun Downstream, Set to Succeeded, Resume (MaxCompute only), Emergency Operation |
Waiting for time | Yes | Stop, Emergency Operation |
Waiting for resource | Yes | Stop, Emergency Operation |
Frozen | Yes | Resume (Unfreeze) |
Not Run | No | Rerun, Rerun Downstream, Emergency Operation |
Succeeded | No | Rerun, Rerun Downstream |
DAG features
Node aggregation
Node aggregation groups nodes in the DAG by a shared dimension, letting you count and compare nodes at a glance instead of inspecting them individually.
Supported aggregation dimensions: node status, workspace, owner, priority.
To aggregate nodes, select an aggregation dimension from the toolbar. The DAG redraws to show group counts for that dimension.
Aggregation by status is available only in DAGs for instances — auto-triggered recurring instances, manually-triggered data backfill instances, and test instances.
The following example compares the same DAG before and after aggregation by priority.
Not aggregated: Each node is displayed individually, showing the full graph structure.
Aggregated by priority: Nodes are grouped by priority level. For example, the current node may show 6 child nodes at priority 1.
Upstream Analysis
Upstream Analysis highlights the ancestor nodes that are blocking a target node from running. Use it when a node has been stuck in a non-running state longer than expected.
Upstream Analysis is supported only for DAGs of instances (auto-triggered recurring instances, manually-triggered data backfill instances, and test instances). It can only be performed on nodes that have not run.
To run an upstream analysis:
Select the target node in the DAG.
Click Upstream Analysis in the upper-left corner.
The DAG highlights ancestor nodes whose state is blocking execution. Blocking states include: running, failed, waiting for time, waiting for resource, and frozen. Up to six ancestor layers are displayed. To display more, click Continue Analysis in the upper-left corner.
Once you locate a blocking node, you can:
Switch to an aggregation view to see the workspace or owner of the blocking instance.
Right-click the blocking instance and select Run Diagnosis to investigate why it has not run. For details, see Intelligent Diagnosis.
Downstream Analysis
Downstream Analysis shows the nodes that are affected by the current node. Use it when you need to assess the impact of a failure or planned change before taking action.
Downstream Analysis is supported only for DAGs of instances (auto-triggered recurring instances, manually-triggered data backfill instances, and test instances).
To run a downstream analysis:
Select the target node in the DAG.
Click Downstream Analysis in the upper-left corner.
After clicking Downstream Analysis, nodes are aggregated by owner by default, and the total number of nodes across all layers is counted. You can filter by dimensions such as node status, workspace, owner, or priority. Up to six descendant layers are displayed. To display more, click Continue Analysis in the upper-left corner.
In a non-aggregated state, if there are more than 10 descendant nodes, they are displayed in groups by default. You can choose how the groups are displayed:
Merge: Merges the analyzed downstream nodes into a single layer based on the current aggregation dimension, showing the count of nodes in each group. For example, when aggregating by workspace, all descendant nodes of the tag node are merged into a single layer showing the number of nodes per workspace.
Layered: Displays the analyzed downstream nodes in layers based on the current aggregation dimension. For example, when aggregating by workspace, the number of nodes per workspace is shown in separate layers.
Click Ungroup to view each node individually and see the full upstream and downstream relationships of all nodes.
Display options
Use the function icons in the upper-right corner of the DAG to adjust display style.
Option | Description |
Full Screen | Expands the DAG to fill the screen. |
Fit to Screen | Scales the DAG to fit within the current viewport. |
Ungroup | Displays upstream and downstream relationships of all nodes individually. |
Group | Groups the display, showing five nodes per row. |
Right-click operations
Right-click any node in the DAG to open the context menu. Available operations depend on the current state of the node. Use the following table as a reference.
Operation | When available | What it does |
Expand Parent/Child Nodes | When a business flow has three or more nodes and some are hidden | Reveals nodes that Operation Center automatically hides. |
View Operational Log | Instances in running, succeeded, or failed state | Opens the task's execution log. Shared resource groups: logs retained one week, instances retained one month. Exclusive resource groups: both retained one month. |
Run Diagnosis | Any instance | Provides full-link diagnostic information. See Intelligent Diagnosis. |
View Code | Any instance | Displays the code of the current instance. |
Edit Node | Any instance | Navigates to the node's edit page in Data Studio. |
View Lineage | Any instance | Displays the data lineage of the current instance. |
View More Details | Any instance | Opens the instance details panel with tabs: Properties, Context, Operational Log, Operation Logs, Code. |
Stop | Instances waiting to run or currently running | Changes instance state to failed. |
Rerun | Instances in Not Run, Succeeded, or Failed state | Reruns the node. After the node succeeds, its descendant nodes that have not run are triggered. |
Rerun Downstream | Instances in Not Run, Completed, or Failed state | Reruns the node and its selected descendants. If a selected node does not meet the run conditions, the error message "The Selected Nodes Include Nodes That Do Not Meet The Run Conditions" is displayed. |
Set to Succeeded | Instances in Failed state only; cannot be used on a business flow | Changes state to succeeded and triggers descendant nodes. |
Resume | After a node fails; supported only for MaxCompute nodes | Resumes the failed node. Note For exclusive resource groups for scheduling purchased before January 2021, to upgrade, you must first click the request link to join the "Alibaba Cloud Big Data & AI Platform" communication group. Then, scan the QR code below to join the DataWorks product DingTalk group for pre-sales and after-sales consultation. |
Emergency Operation > Remove Dependency | Any instance | Removes the dependency of the current node. Valid only for the current run. |
Pause (Freeze) | Running instances only | Freezes the current instance only. A frozen instance is not auto-scheduled and blocks its descendant nodes. Freezing one instance does not affect sibling instances. For example, if a node runs hourly (24 instances per day), freezing one instance does not affect the other 23 instances for that day. |
Resume (Unfreeze) | Frozen instances | Unfreezes the current instance only. If the instance has not yet run, it runs automatically after ancestor nodes complete. If all ancestors have already completed, the instance is set to failed and must be manually rerun. The next day's auto-triggered instance remains frozen. |
Instance details panel
Click an instance in the DAG to open the details panel in the lower-right corner. Click Expand Details to view the full panel.
The panel contains the following tabs.
Tab | What it shows |
Properties | Scheduling properties of the task in the production environment. See Scheduling Configuration. Key information includes: node ID (locates all hourly or minute-level instances for that node on the current day) vs. instance ID (locates a specific instance). The tab also shows rule monitoring and baseline associations — click Create to quickly add a new monitoring rule or baseline. |
Context | Contextual information for the current instance. |
Operational Log | The task's execution log — a record of what the engine did during the run. |
Operation Logs | The audit trail — a record of manual actions taken on the task or instance (time, operator, and action performed). |
Code | The latest code of the current task in the production environment. If the code is not as expected, verify that the latest task version has been published. See Publish a task. |
Operational Log records what the system executed. Operation Logs records who did what manually.
On the Properties tab, the following troubleshooting guidance applies:
Task status: If a task is in Not Run, Waiting for Time, Waiting for Resource, or Frozen state, use Run Diagnosis to quickly locate the issue.
Waiting for resources: If a task waits for resources for a long time, use Run Diagnosis to identify which tasks are occupying resources. This helps you quickly find and troubleshoot abnormal tasks.
Excessive runtime: If a task's runtime is significantly longer than its historical runtimes, handle it based on the task type:
For non-sync tasks, consult the owner of the corresponding engine.
For offline sync tasks, a certain stage may be running slowly or waiting for resources. See Batch synchronization.
Rule monitoring: View the monitoring rules associated with the current instance. Click Create to quickly add a rule for the task's running status. See Rule management. Note: Only rules that monitor task running status are shown here; Data Quality monitoring rules are not shown.
Baseline monitoring: View the baselines associated with the current instance. Click Create to quickly create a baseline. See Baseline management.
What's next
Intelligent Diagnosis — Run a full-link diagnosis on a stuck or failed instance.
Scheduling Configuration — Review the scheduling properties visible in the Properties tab.
Publish a task — Deploy the latest code to the production environment.
Software edition selection and billing — Upgrade to Professional Edition to unlock Upstream Analysis, Downstream Analysis, and node aggregation.