June 12, 2026 release - Realtime Compute for Apache Flink

Realtime Compute for Apache Flink introduces expanded AI capabilities, Agent Skill publishing, and auto-tuning improvements on June 12, 2026.

Important

This upgrade will roll out gradually across all regions. For the detailed rollout schedule, check the latest announcements on the right side of the Realtime Compute console. If you cannot access the new features, your account has not yet been included in the grayscale rollout. To request an early upgrade, or submit a ticket, and we will arrange it based on actual conditions.

Overview

This release expands AI-powered capabilities and improves the auto-tuning experience. The AI assistant now invokes Skills for job management, fault diagnosis, and resource tuning, letting you perform common O&M operations through natural language. Flink Skills are listed in Understanding Agent Skills, enabling external agents such as DataWorks to install and integrate them on demand for cross-product natural language control of Flink. AI Function now supports service registration through Alibaba Cloud Model Studio account packages, removing the need to provide your own API key. Auto-tuning adds support for Flink CDC YAML data ingestion jobs and integrates with health scores for better visibility into tuning status.

Platform side

AI capabilities

AI assistant gains Skill invocation capability: The console AI assistant can trigger Skills—such as job operations, O&M diagnostics, resource tuning, and Q&A—through natural language. It returns structured results with parameter recommendations and next-step guidance.
First batch of Agent Skills published to Alibaba Cloud Skill Hub: Realtime Compute for Apache Flink publishes its first two Agent Skills on the Alibaba Cloud Skill Hub. These Skills integrate with AI Agent platforms such as Claude Code and OpenClaw, enabling natural language–based management of Flink workspace instances and job O&M operations.
- alibabacloud-flink-instance-manage: Manages the lifecycle of Flink instances and namespaces. Supports querying available regions and zones, creating instances, checking instance status, creating and querying namespaces, and listing resource tags. Ideal for instance provisioning and capacity planning.
- alibabacloud-flink-workspace-ops: Handles day-to-day workspace O&M tasks. Covers full lifecycle management (start, stop, diagnose, delete) for Session Clusters and deployed jobs, plus workspace configuration operations such as table queries, member management, and variable management.

Platform features

New purchases support pvtz in shared VPC scenarios: New Flink instances can now be purchased in shared VPC environments without additional network configuration.
Workspace access permissions by resource group: Workspace access can be controlled at the Alibaba Cloud resource group level for multi-team isolation.
Billing console integrates Alibaba Cloud resource group authorization: Supports access control at the resource group level.

Experience improvements

Auto-tuning:

Adds support for Flink CDC YAML data ingestion jobs: Auto-tuning now covers data ingestion scenarios.
Supports dynamic TM CPU scaling: Adjust TaskManager CPU allocation without restarting the job.
Auto-tuning integrates with health scores: When health score diagnostics detect resource-related threats, auto-tuning triggers automatically, closing the loop from issue detection to one-click tuning.

Resource queues:

Optimizes used CU calculation logic: When the CPU-to-memory ratio is not 1:4, used CUs are calculated as max(CPU-equivalent CUs, memory ÷ 4), taking the larger value for more accurate display and preventing underestimated resource utilization.

OpenAPI

Adds job health score (GetJobDiagnosis): Returns health score data, enabling external systems to retrieve real-time job health status.