Scheduling dependency rules and examples-Dataphin(Dataphin)-阿里云帮助中心

Background information

Before you configure scheduling dependencies, understand the following concepts.

In Dataphin, a recurring task generates instances according to its scheduling cycle. For example, an hourly task generates a set of hourly instances each day. A task runs by executing these instances, and a scheduling dependency between tasks is essentially a dependency between their instances. Instance dependencies vary when the upstream and downstream tasks have different scheduling cycles.
Dataphin supports various scheduling dependency scenarios. A task can depend on the root node, the previous cycle of the same node, or other physical/logical table nodes.
Conditions for an instance to start running

An instance enters the Waiting for Scheduling Time or Waiting for Scheduling Resources state after its dependent upstream instances run successfully. The instance status changes to Running and the task begins to execute when its scheduled time is reached, scheduling resources are allocated, and the system is not throttling. An instance for a trigger-based node also requires an external trigger request.
A downstream instance will not start until its dependent upstream instance has run successfully, even if the downstream instance's scheduled time has passed. The scheduled time is the planned time, not the actual execution time.
At 23:00 every night, the system automatically generates the recurring instances for the next day. If a recurring task has no instances scheduled to run on a given day, the system creates a dummy instance and immediately marks it as Succeeded. For example, for a monthly task scheduled to run at 13:00:00 on the second day of each month, the system will generate a dummy instance and mark it as Succeeded on all other days of the month.

Overview

When you configure a scheduling dependency, you can preview it in the Node Dependency Cycle Preview pane of the Edit Dependency dialog box. For more information, see Configure scheduling dependencies for an offline task.

The following table describes the sections in this preview pane.

Section	Description
①	Lists the instances for the current node. Click an instance to select it; the selected instance is highlighted in blue.
②	Lists instances for the selected upstream node. Upstream instances required by the selected downstream instance are highlighted in blue.
③	Lines connect the selected downstream instance (right) to its dependent upstream instances (left).

Dependency scenarios

The following sections describe dependency rules for recurring tasks with different scheduling cycles.

Note

If a downstream task is scheduled earlier than its upstream task, it will not run at its scheduled time. Instead, it will wait for the upstream task to complete.

Dependency on the root node

When a task depends on the root node, no dependency cycle or policy can be configured. The root node is always considered to have run successfully, so the task runs at its scheduled time once scheduling resources are allocated. The root node does not block downstream nodes.

Dependency on other nodes or previous cycles

When a task depends on the previous cycle of the same node or on a physical/logical table node, you can configure a dependency cycle and a dependency policy.

dependency cycle: Options include current cycle (same day), previous cycle (previous day), N days ago, and last 24 hours.

Note
When setting a dependency on the same node, you cannot select current cycle (same day). A node cannot have a same-cycle dependency on itself.
dependency policy: Options include All Instances, First Instance, Last Instance, next nearest instance, and previous nearest instance. You can also specify the required status of the upstream instances: Succeeded, Failed, or Complete (Succeeded or Failed). The following examples illustrate each dependency policy using a typical scenario: an hourly task that depends on another hourly task, which must have a status of Succeeded.

Dependency cycle	Dependency policy
current cycle (same day)	When the dependency cycle is set to current cycle (same day), the dependency policies work as follows: All Instances For an All Instances dependency, all instances of the upstream task for the current cycle (same day) must run successfully before the downstream nodes can run normally. For example, if the business date is `2024-03-02`, the dependency task for the current cycle (same day) is the task instance for `2024-03-02`. The execution of the instance nodes for the current task depends on all instances of the `2024-03-02` dependency node. If all instances of the `2024-03-02` dependency node run successfully, all instance nodes of the current task for `2024-03-02` can run normally. If an instance of the `2024-03-02` dependency node fails to run successfully, all instance nodes for the `current task 2024-03-02` cannot run normally. Example in the Node Dependency Cycle Preview: First Instance For the First Instance, subsequent nodes can run normally as long as the first instance of the dependency task in the current cycle (same day) succeeds. For example, if the business date is `2024-03-02`, the current cycle (same day) is also `2024-03-02`. The instance nodes of the current task for `2024-03-02` depend on the first instance of the dependency node for `2024-03-02`. If the first instance of the dependency node for `2024-03-02` succeeds, all instance nodes of the current task for `2024-03-02` can run normally. If the first instance of the dependency node for `2024-03-02` does not succeed, none of the instance nodes of the current task for `2024-03-02` can run normally. Example of the dependency cycle preview: Last Instance For the Last Instance setting, a downstream node can run only after the last instance of its upstream task in the current cycle (same day) has Succeeded. For example, if the data timestamp is `2024-03-02`, the current cycle (same day) is also `2024-03-02`. The instance of the current task for `2024-03-02` depends on the last instance of the upstream node for `2024-03-02`. If the last instance of the upstream node for `2024-03-02` has Succeeded, all instances of the current task for `2024-03-02` can run. If the last instance of the upstream node for `2024-03-02` has not Succeeded, all instances of the current task for `2024-03-02` cannot run. Example of the dependency cycle preview: Next nearest instance This is the first instance within the selected interval that runs at or later than the current instance. If no matching instance exists, the Last Instance is used. For example, if the business date is `2024-03-02`, the current cycle (same day) is also `2024-03-02`. For a task instance node with a business date of `2024-03-02` and a scheduled time of `13:05:00`, the next nearest instance is the instance scheduled at `14:00:00` on `2024-03-02`. This means the `13:05:00` instance node of the current task on `2024-03-02` depends on the `14:00:00` instance of the dependent node in the current cycle (same day). For a task instance node with a business date of `2024-03-02` and a scheduled time of `13:00:00`, the next nearest instance is the instance scheduled at `13:00:00` on `2024-03-02`. This means the `13:00:00` instance node of the current task on `2024-03-02` depends on the `13:00:00` instance of the dependent node in the current cycle (same day). The same logic applies to other nodes. Example of the dependency cycle preview: Previous nearest instance This is the instance in the selected range with a scheduled time closest to the current task's scheduled time. The system gives priority to the last instance with a scheduled time earlier than the current task. For example, if the business date is `2024-03-02`, the current cycle (same day) is also `2024-03-02`. For a task instance node with a business date of `2024-03-02` and a scheduled time of `13:05:00`, the previous nearest instance is the instance scheduled at `13:00:00` on `2024-03-02`. This means the `13:05:00` instance node of the current task on `2024-03-02` depends on the `13:00:00` instance of the dependency node in the current cycle (same day). For a task instance node with a business date of `2024-03-02` and a scheduled time of `13:00:00`, the previous nearest instance is the instance also scheduled at `13:00:00` on `2024-03-02`. This means the `13:00:00` instance node of the current task on `2024-03-02` depends on the `13:00:00` instance of the dependency node in the current cycle (same day). The same logic applies to other nodes. The following images show a preview of the node dependency cycle when you configure a scheduling dependency.
previous cycle (previous day)	When the dependency cycle is set to previous cycle (previous day), the dependency policies work as follows: All Instances For the All Instances setting, all instances from the previous cycle (previous day) must run successfully before dependent downstream nodes can run. For example, if the business date is `2024-03-02`, the previous cycle (previous day) is `2024-03-01`. The instance nodes for `2024-03-02` depend on all instances from `2024-03-01`. If all instances from `2024-03-01` run successfully, the instance nodes for `2024-03-02` can run normally. If even one instance from `2024-03-01` fails to run successfully, none of the instance nodes for `2024-03-02` can run normally. Example of the dependency cycle preview: First Instance For the First Instance dependency, if the first instance of the previous cycle (previous day) succeeds, all dependent downstream nodes can run. For example, if the current business date is `2024-03-02`, the previous cycle (previous day) is `2024-03-01`. The instance for `2024-03-02` depends on the first instance of `2024-03-01`. If the first instance for `2024-03-01` succeeds, all instances for `2024-03-02` can run. If the first instance for `2024-03-01` does not succeed, no instances for `2024-03-02` can run. Example of the dependency cycle preview: Last Instance With the Last Instance option, all subsequent dependent nodes run only if the last instance from the previous cycle (previous day) succeeded. For example, if the current business date is `2024-03-02`, the previous cycle (previous day) is `2024-03-01`. The instance nodes for `2024-03-02` depend on the last instance from `2024-03-01`. If the last instance from `2024-03-01` succeeded, all instance nodes for `2024-03-02` run normally. If the last instance from `2024-03-01` did not succeed, all instance nodes for `2024-03-02` do not run. Example of the dependency cycle preview: Next nearest instance This is the first instance in the selected range that runs later than the current instance. If no matching instance exists, the Last Instance is used. For example, if the business date is `2024-03-02`, the previous cycle (previous day) is `2024-03-01`. For a task instance node with a business date of `2024-03-02` and a scheduled time of `13:00:00`, the next nearest instance is the instance with a business date of `2024-03-01` and a scheduled time of `23:00:00`. This means the `13:00:00` instance of the current task on `2024-03-02` depends on the `23:00:00` instance of the dependency node from the previous cycle (previous day). The same logic applies to other nodes. Example of the dependency cycle preview: Previous nearest instance The instance within the selected interval whose scheduled time is closest to the scheduled time of the current task. Priority is given to the last instance that is scheduled before the current task. For example, if the data timestamp for the current day is `2024-03-02`, the previous cycle (previous day) is `2024-03-01`. For a task instance node with a data timestamp of `2024-03-02` and a scheduled time of `13:00:00`, the previous nearest instance is the instance with a data timestamp of `2024-03-01` and a scheduled time of `13:00:00`. As a specific example, the `13:00:00` instance node of the current task on `2024-03-02` depends on the `23:00:00` instance of the dependent node from the previous cycle (previous day). The same logic applies to other nodes. The following image shows an example of the dependency cycle preview when you configure a scheduling dependency.
N days ago	When the dependency cycle is set to N days ago, the logic varies by policy. The following examples use 2 days ago to illustrate each dependency policy: All Instances For All Instances, dependent subsequent nodes can run normally only after all instances from two days prior have run successfully. For example, if the current business date is `2024-03-04`, 'two days prior' is `2024-03-02`. The instance nodes for `2024-03-04` depend on all instances from `2024-03-02`. If all instances from `2024-03-02` run successfully, all instance nodes for `2024-03-04` can run normally. If even one instance from `2024-03-02` does not run successfully, none of the instance nodes for `2024-03-04` can run normally. The same logic applies to other nodes. Example of the dependency cycle preview: First Instance With the First Instance option, subsequent dependent nodes can run as expected only if the first instance from two days prior has succeeded. For example, if the current business date is `2024-03-04`, then two days prior is `2024-03-02`. The instance nodes for `2024-03-04` depend on the first instance of `2024-03-02`. If the first instance of `2024-03-02` has succeeded, all instance nodes for `2024-03-04` can run as expected. If the first instance of `2024-03-02` has not succeeded, all instance nodes for `2024-03-04` cannot run as expected. Example of the dependency cycle preview: Last Instance For the Last Instance dependency, subsequent nodes can run only if the last instance from two days prior succeeded. For example, if the current business date is `2024-03-04`, then two days prior is `2024-03-02`. An instance for `2024-03-04` depends on the last instance from `2024-03-02`. If the last instance from `2024-03-02` succeeded, all instances for `2024-03-04` can run. If the last instance from `2024-03-02` did not succeed, no instances for `2024-03-04` can run. Example of the dependency cycle preview: Next nearest instance This refers to the instance within the selected range whose scheduled time is closest to the current instance's scheduled time. If no matching instance exists, the Last Instance is used. For example, consider a dependency set to two days prior. If the current business date is `2024-03-04`, the dependency is on an instance from `2024-03-02`. The next nearest instance for a task scheduled at `13:00:00` on `2024-03-04` is the instance scheduled at `13:00:00` on `2024-03-02`. This means the `13:00:00` instance on `2024-03-04` depends on the `13:00:00` instance of the dependent node from two days prior. The same logic applies to other instances. Example of the dependency cycle preview: Previous nearest instance This is the instance in the selected range whose scheduled time is closest to the current task's scheduled time. The last instance scheduled to run before the current task has priority. For example, a task depends on an instance from two days ago. The current data timestamp is `2024-03-04`, so the dependency date is `2024-03-02`. For a task instance scheduled at `13:00:00` on `2024-03-04`, the previous nearest instance is the one scheduled at `13:00:00` on `2024-03-02`. Therefore, the instance at `13:00:00` on `2024-03-04` depends on the upstream node's instance at `13:00:00` from two days prior. This logic applies to other nodes. Example of the dependency cycle preview:
last 24 hours	When the dependency cycle is last 24 hours, each downstream instance depends on upstream instances within the 24-hour window immediately preceding its scheduled time. The specific behavior is determined by the dependency policy: All Instances For All Instances, the current node can run as normal only if all its dependent instances from the last 24 hours have run successfully. For example, if the business date is `2024-03-04`, the dependent instances within 24 hours of the `13:00:00` instance include the `13:00:00`, `14:00:00`, and `15:00:00` instances from `2024-03-03`. The dependent instances within 24 hours of the `14:00:00` instance include the `14:00:00` and `15:00:00` instances from `2024-03-03`. Example of the dependency cycle preview: First Instance With the First Instance setting, the current node runs only after the first dependent instance from the last 24 hours runs successfully. For example, if the business date is `2024-03-04`, the dependent instances for the `13:00:00` instance in the last 24 hours include `2024-03-03`'s `13:00:00`, `14:00:00`, and `15:00:00` instances. The first instance in this range is `13:00:00`. For the `14:00:00` instance, the dependent instances in the last 24 hours include `2024-03-03`'s `14:00:00` and `15:00:00` instances. The first instance in this range is `14:00:00`. The following image shows an example of the dependency cycle preview when you configure a scheduling dependency. Last Instance With the Last Instance setting, the current node runs only after the last dependent instance from the last 24 hours has run successfully. For example, if the business date is `2024-03-04` and the scheduled time of the current instance is `13:00:00`, the time range for dependent instances in the last 24 hours is `[2024-03-03 13:00 to 2024-03-04 13:00)`. Example of the dependency cycle preview: Next nearest instance This is the most recent instance within the selected interval that runs before the current instance's runtime. If no matching instance exists, the Last Instance is used. For example, the business date is `2024-03-04`. When the current instance's scheduled time is `2024-03-04 13:00`, the search range for the next nearest instance is `[2024-03-03 13:00 to 2024-03-04 13:00 )`, and the actual dependent instance is `2024-03-04 12:00`. When the current instance's scheduled time is `2024-03-04 13:05`, the search range is `[2024-03-03 13:05 to 2024-03-04 13:05 )`, and the actual dependent instance is `2024-03-04 13:00`. Example of the dependency cycle preview: Previous nearest instance This is the instance within the selected range whose scheduled time is closest to the scheduled time of the current task. Priority is given to the last instance that is scheduled to run before the current task. For example, if the data date is `2024-03-04` and the current instance is scheduled for `2024-03-04 13:00`, the search range is `[2024-03-03 13:00 to 2024-03-04 13:00)`. The actual dependent instance is the one scheduled at `2024-03-04 12:00`. If the current instance is scheduled for `2024-03-04 13:05`, the search range is `[2024-03-03 13:05 to 2024-03-04 13:05)`, and the actual dependent instance is the one scheduled at `2024-03-04 13:00`.