Platform For AI:Configure an update plan for an EAS service

Key concepts

• Multiple update modes: Two modes are supported — manual batching (Partition) and automatic batching (Batch Update).

  Characteristic	Manual batching (Partition)	Automatic batching (Batch Update)
  Control	Fully manual; you advance each batch	System advances batches automatically after configuration
  Operational overhead	Higher	Lower
  Flexibility	Highest	Moderate
  Best for	Canary releases of critical online services	Routine updates to large-scale services

• Full control throughout the update: Pause, roll back, or change the batching strategy at any point during the update. Delete the plan to revert to the default full-replacement strategy.

• Real-time status visibility: An Update plan status tag in the service details header shows the current phase. Hover over the tag to see a tooltip with live progress, batch countdown, and other key information.

Quick start

The following example walks through creating an update plan and completing a service update using manual batching on a service with 10 replicas.

Prerequisites: enable the update plan feature

When creating a service, add "eas.aliyun.com/enable-rollout": "true" to the features field of the JSON configuration. Without this setting, the update plan feature is unavailable. Example:

{
  "metadata": {
    "instance": 10
  },
  "features": {
    "eas.aliyun.com/enable-rollout": "true"
  }
}

Step 1: Create an update plan

1. Click the target service to open the EAS service details page.

2. Locate the Update plan tag in the header bar. By default it appears gray, indicating no plan is configured.

3. Hover over the tag and click Create update plan in the tooltip that appears.

4. In the panel that slides in from the right, select Manual batching and set the first-batch replica count to 2.

5. Click OK.

The update plan is created. The tag turns blue and the status changes to Pending.

Step 2: Trigger the update

Click Update on the service details page and submit the new service configuration.

The system starts the update according to the plan: it updates 2 replicas first, then pauses and waits. The tooltip shows the current progress (2/10).

Step 3: Verify and advance to the next batch

Check your business metrics (latency, error rate, and so on). If no issues are found, modify the plan and set the target replica count to 10. The system updates the remaining 8 replicas until the update is complete.

Status reference

The update plan uses a two-level status model: the top level is the phase, and the second level is the stage (which only exists when the phase is Active).

• Not configured (gray): No update plan exists. Updating the service triggers a full replacement.

• Pending (blue): A plan has been created and is waiting for a service update to activate it.

• Active (blue): The update has been triggered and the system is replacing replicas in batches. Three stages exist within Active:

  • Current batch executing: The batch is in progress; the updated count is approaching the target.

  • Current batch paused: A user paused execution; progress is frozen.

  • Current batch completed: The batch target has been reached. The system waits for you to advance manually (manual batching) or for the countdown to expire (automatic batching).

• Completed (blue): The current round of updates is finished.

  • Updating the service again re-triggers the plan and changes the status to Active.

  • Modifying the plan changes the status to Pending and waits for the next service update.

The following table shows which operations are available in each status:

Usage notes

• Modifying an Active plan: New parameters take effect immediately. If the plan is currently paused, the modification automatically resumes it.

• Modifying a Completed plan: The configuration is saved but does not trigger an update automatically. Update the service again to apply the new plan.

• Deleting an Active plan: The remaining replicas continue updating using the full-replacement strategy. No rollback occurs.

• Scaling while an update plan exists: The plan parameters remain unchanged and apply based on the actual replica count at execution time.

• Setting the first-batch count to 0 in manual batching: This is a valid value. The service update triggers immediately but pauses at the starting point, waiting for you to advance manually.

• Setting the batch size ≥ total replica count in automatic batching: All replicas are updated in a single batch, which is equivalent to a full replacement.

JSON configuration

Configure the update plan by setting parameters in the metadata.rolling_strategy field of the service JSON configuration.

{
  "metadata": {
    "rolling_strategy": {
      "max_surge": "25%",
      "max_unavailable": "20%",
      "partition": 5,
      "paused": false,
      "batch_update": {
        "interval": "5m",
        "batch_size": 2
      }
    }
  }
}

{
  "metadata": {
    "instance": 10,
    "rolling_strategy": {
      "partition": 2,
      "max_unavailable": 1
    }
  }
}

{
  "metadata": {
    "instance": 100,
    "rolling_strategy": {
      "batch_update": {
        "interval": "10m",
        "batch_size": 5
      }
    }
  }
}

{
  "metadata": {
    "rolling_strategy": {
      "paused": true
    }
  }
}

{
  "metadata": {
    "rolling_strategy": {
      "partition": 0
    }
  }
}

FAQ