This page defines the core terms and concepts you need to design and build CloudFlow workflows.
Key terms at a glance
| Term | Description |
|---|---|
| CloudFlow | A service that coordinates, manages, and executes distributed tasks through visual workflow orchestration. |
| Standard mode | Workflow mode for complex, long-running orchestration with persistent step-status storage. |
| Express mode | Workflow mode for latency-sensitive, load-intensive orchestration with fast deployment. |
| Flow Definition Language (FDL) | JSON-based language for defining state machines that drive workflow execution. |
| State | An individual step in a state machine that performs work, controls flow, or handles errors. |
| Workflow integration | Direct calls to Alibaba Cloud service APIs from within a workflow. |
| Data passing | The mechanism that transfers data between states during workflow execution. |
| Error handling | Built-in Retry and Catch mechanisms for managing failures during execution. |
| Integration mode | The pattern that controls how a Task state interacts with an integrated service: RequestComplete, WaitForSystemCallback, or WaitForCustomCallback. |
CloudFlow
CloudFlow coordinates, manages, and executes multiple distributed tasks. Visually organize services, API calls, and data processing steps into an ordered workflow. Debug each step to verify the workflow runs as expected.
To get started, see Create a workflow.
Workflow modes
CloudFlow offers two execution modes. Choose the mode that matches your workload.
| Dimension | Standard mode | Express mode |
|---|---|---|
| Best for | Complex logic, long-running offline orchestration | Low-latency, load-intensive online orchestration |
| Step-status persistence | Yes -- stores every step's status | No |
| Typical use cases | Multi-step approval flows, batch data pipelines, workflows with complex branching | Microservice API orchestration, glue-layer API orchestration, quasi-real-time workflow orchestration, stream processing |
For a detailed comparison, see Standard mode and express mode.
Standard mode
Standard mode persists the status of every execution step. Configure each step individually for complex logic judgments and conventional offline workflow orchestration that require status tracking across steps.
Express mode
Express mode is suitable for simple structures and solid processes, supporting fast deployment. It targets latency-sensitive, load-intensive workloads such as microservice API orchestration, glue-layer API orchestration, quasi-real-time workflow orchestration, and stream processing.
Workflow integration
CloudFlow integrates with most Alibaba Cloud service APIs through two integration types:
Normal integration
CloudFlow calls Alibaba Cloud service APIs directly, without extra packaging or optimization. Normal integration supports up to 10,000 API operations across Alibaba Cloud services.
For more information, see Normal integration.
Optimized integration
Frequently used API operations are pre-packaged to simplify calls and service integration. For services or tools not available through OpenAPI, use their SDKs or public clients through optimized integration to access their capabilities within workflows.
For more information, see Overview of workflow integration.
Flow Definition Language (FDL)
Flow Definition Language (FDL) is a JSON-based language for defining state machines. A state machine is a collection of states. Each state performs one action: doing work (Task), choosing the next state (Choice), or stopping on error (Fail).
Scopes
States can be combined and nested. When one state contains another, the outer state is the scope of the inner state.
For more information, see the Scopes of states section of the "Flow definition" topic.
States
States are the building blocks of a state machine. Each state represents an execution stage and determines what happens next. CloudFlow provides eight state types in two categories:
Work states
Work states perform actions or transform data.
| State | Purpose |
|---|---|
| Task | Call an integrated service API -- invoke functions, Alibaba Cloud service APIs, or third-party services over HTTP or HTTPS. |
| Pass | Pass input directly to output without performing work. Use as a blank node or as a data-preprocessing node to transform input into the expected output format. |
| Wait | Pause execution for a specified duration, then resume. |
Flow-control states
Flow-control states determine execution order and branching.
| State | Purpose |
|---|---|
| Choice | Branch execution based on conditions. Each branch contains a Condition field (conditional expression) and a Next field (target state). A default branch handles unmatched conditions. |
| Parallel | Execute multiple branches of child states concurrently. |
| Map | Iterate over an array parameter in the input and execute the Processor states in parallel for each element. |
| Succeed | Terminate a series of states before the states are executed. |
| Fail | Handle errors and terminate a series of states before the states are executed. |
For more information: Choice | Parallel | Map | Succeed | Fail
Data passing
Data passing is the process of transferring data between states during execution. The data falls into two categories:
Workflow data: metadata such as the workflow name, region, and the names of its constituent states.
Execution data: runtime data such as the execution name, execution input, and the input and output of the current state.
For more information, see Data passing.
Inputs and outputs
Each workflow execution accepts a JSON object as input and passes it to the first state. Every subsequent state receives a JSON input, processes it, and passes a JSON output to the next state.
Execution input and output: the data in the overall execution context, covering inputs and outputs of all states across the workflow run.
State input and output: the data scoped to a single state, such as the parameters it receives and the results it produces.
For more information, see Inputs and outputs.
Built-in functions
FDL provides built-in functions for basic data-processing operations within workflow definitions. Use them to reduce code compilation workloads and shorten the development cycle, helping you quickly deploy and update workflows.
For more information, see Built-in functions.
Error handling
CloudFlow provides two complementary error-handling mechanisms: Retry and Catch.
Retry
When a step fails and the corresponding state includes a Retry field, CloudFlow automatically retries the step according to the specified retry policy. The step may succeed after a few retries.
Catch
If a step still fails after all retries, a Catch block routes execution to error-handling logic -- such as sending notifications or performing rollback operations -- to ensure system stability and data integrity.
Together, Retry and Catch let you manage and handle various errors that occur during workflow execution in an efficient manner, improving the robustness and reliability of the system. For syntax details and examples, see Error handling.
Integration modes
Integration modes control how a Task state interacts with the service it calls. CloudFlow supports three modes:
| Mode | Behavior |
|---|---|
| RequestComplete | Default. CloudFlow sends the request and transitions to the next state after receiving a response. |
| WaitForSystemCallback | CloudFlow sends the request and waits for the integrated service to report task completion before transitioning. |
| WaitForCustomCallback | CloudFlow sends the request with a task token, then suspends execution. You can control the execution of the workflow based on the logic that you defined in the integration service. |
For more information, see Service integration modes.