All Products
Search

This Product

    Realtime Compute for Apache Flink:Start a deployment

    Document Center

    Realtime Compute for Apache Flink:Start a deployment

    Last Updated:May 27, 2026

    After developing and deploying a deployment, you must start it. You also need to start a deployment to resume it after it has been stopped or to apply configuration changes that require a restart.

    Prerequisites

    You have created a deployment. For more information, see Deploy a deployment.

    Limitations

    Startup options are available only for streaming deployments.

    Precautions

    • If a RAM user, RAM role, or another Alibaba Cloud account starts a deployment, make sure the account has the required permissions to access the target namespace. For more information, see Authorize access in the console and Manage permissions.

    • When you start a deployment from its latest state or a specific state, the system performs a state compatibility check. Starting a deployment that has state compatibility issues can lead to startup failures or unexpected results. For more information, see Flink state compatibility reference.

    Procedure

    1. Navigate to the Deployments page.

      1. Log on to the Realtime Compute for Apache Flink console.

      2. In the top navigation bar, select the target namespace.

      3. In the O&M > Deployments pane, select STREAM or BATCH from the drop-down list.

    2. Find the target deployment and click Start in the Actions column.

    3. (Optional) For a streaming deployment, configure the startup options.

      • Initial mode

        Use this mode to launch a new deployment or to start a deployment without using a previous state.

        Strategy

        Description

        Specify source's start time

        Select Specify source's start time and specify a time.

        You can set the source start time for Kafka, SLS, DataHub, ApsaraMQ for RocketMQ, Hologres, Paimon, and MySQL connectors.

        The start time specified on the deployment start page overrides the startTime parameter set in the deployment's DDL code.

        Note

        • Kafka versions earlier than 0.11 might not be supported due to potential incompatibility issues with the connector's client version. We recommend upgrading your Kafka version.

        • Not all connectors support the startTime parameter. To check for support, see if the WITH clause for a specific connector includes the startTime parameter. For an example, see SLS WITH parameters.

        • The startTime parameter takes effect only for new deployments started with this option. It is ignored if you resume a deployment from a system checkpoint or a savepoint.

        Configure automatic tuning

        After you enable this option, select a tuning mode:

        • Autopilot Mode: The system automatically scales down resources when usage is low and scales them up when usage exceeds a certain threshold. For more information, see Enable and configure Autopilot Mode.

        • Scheduled Mode: Select a schedule from the drop-down list. A schedule can contain multiple resource configurations mapped to different times. You can configure resources based on usage patterns for each time period. For more information, see Configure and apply a scheduled tuning plan.

      • Resume mode

        Select a startup strategy.

        Strategy

        Description

        Resume from latest state

        Resumes the deployment from the latest savepoint or system checkpoint. Flink detects changes to the SQL code, Flink runtime parameters, and engine version.

        If changes are detected, click Click to detect next to State Compatibility. For more information about the results and recommended actions, see Compatibility.

        Resume from specified state

        Select a specific savepoint to resume from. For information about how to create a savepoint, see Manage deployment state.

        Resume from another deployment

        Select this option to specify the target deployment and its corresponding savepoint to resume from. You can share a savepoint between deployments, but the deployments must be state-compatible. For more information, see Manage deployment state.

        Allow Non-restored State

        Note

        This option is supported only for JAR deployments.

        By default, Flink tries to map all state from the savepoint to the new deployment's operators. If the deployment has been modified, this mapping can fail. Enabling this option allows Flink to start the deployment by skipping any state that cannot be mapped. For more information, see Allow None-Restored State.

        Configure automatic tuning

        After you enable this option, select a tuning mode:

        • Autopilot Mode: The system automatically scales down resources when usage is low and scales them up when usage exceeds a certain threshold. For more information, see Enable and configure Autopilot Mode.

        • Scheduled Mode: Select a schedule from the drop-down list. A schedule can contain multiple resource configurations mapped to different times. You can configure resources based on usage patterns for each time period. For more information, see Configure and apply a scheduled tuning plan.

    4. Click Start.

      On the O&M > Deployments page, view the deployment status. For more information, see View deployment status.

    Related documentation

    • After a deployment starts, you can modify its runtime parameters. For more information, see Configure runtime parameters. Some parameters also support dynamic updates, which reduces service downtime by avoiding a restart. For more information, see Dynamic scaling and parameter updates.

    • Once a deployment is running, you can track its data lineage to locate issues or assess impact. For more information, see View data lineage.

    • To learn about the enterprise-grade state backend GeminiStateBackend and its performance comparison with RocksDBStateBackend, see Introduction to GeminiStateBackend.