This page lists the feature releases and updates for Function Compute (FC) in 2025.
December 2025
| Feature | Change type | Description | References |
|---|---|---|---|
Apsara DevOps Flow integration with Function Compute 3.0 | New | Function Compute integrates with Apsara DevOps pipelines to enable automatic deployment of functions. | |
Rush mode | New | To resolve the cold start issue that occurs when a function is invoked for the first time in the default mode (auto-scaling by request), Function Compute introduces Rush mode. Rush mode supports startup snapshots. It ensures efficient cold starts while allowing you to lock scarce elastic resources at a low cost. This guarantees delivery for your business and is suitable for latency-sensitive scenarios. | |
AgentRun | New Product Releases | AgentRun for Function Compute is a one-stop Agentic AI infrastructure platform. It features a code-centric approach, an open ecosystem, and flexible assembly. AgentRun provides full lifecycle management for enterprise-level Agentic applications, from development and deployment to O&M. | |
AIO Sandbox (All-in-One Sandbox) | New | AIO Sandbox is a unified, cloud-based isolated environment that integrates BrowserTool and Code Interpreter. It provides one-stop capabilities for web automation, complex calculations, file processing, and interactive terminals. | |
| GPU hybrid instance mode (Shenzhen and Shanghai) | New | Combines resident and on-demand instances. Resident instances handle stable baseline traffic. On-demand instances scale up for peak loads, then enter a low-cost idle state when traffic drops. | |
| Request-based auto scaling for all GPU types | New | All GPU types now scale based on concurrent request volume. The system creates and releases instances dynamically to match your workload. New instances experience cold starts — configure provisioned instances for latency-sensitive services. | |
| GPU image size limit increased to 30 GB | Optimization | The maximum compressed image size for GPU function deployments increased from 15 GB to 30 GB, improving deployment success rates for images with large dependencies or large language models (LLMs). | Quotas and limits |
| PolarDB dynamic mounting (Hong Kong (China), US West, and Singapore) | New | Dynamically mount PolarDB storage directories so different instances can share persistent data and large model weight files. | |
| OSS dynamic mounting (Hong Kong (China), US West, and Singapore) | New | Mount and unmount Object Storage Service (OSS) bucket directories at runtime without restarting services or rebuilding images. Decouples compute from large-scale unstructured data, simplifying model training data loading and multimedia content streaming. | |
| Native Cross-Origin Resource Sharing (CORS) policy for HTTP triggers | New | Configure CORS rules directly on HTTP triggers. Specify allowed origins, request methods, headers, and preflight cache durations to handle cross-domain browser calls without custom middleware. | |
| Console overview page upgrade and FunctionQ assistant | Optimization | The FC overview page is redesigned with Cost Center, Resource Management, and Solutions+AI sections. FunctionQ, an intelligent development assistant, provides document search and technical Q&A for developers. |
November 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| GPU integration with Application Real-Time Monitoring Service (ARMS) for LLM observability | New | Deploy an OpenTelemetry Collector in sidecar mode within a function instance to collect and report Prometheus metrics from the LLM inference engine, including token throughput and request latency. | Integration solution for monitoring LLM inference model service metrics |
| DevPod support for custom image runtimes | New | DevPod is a cloud-based container development environment that integrates VS Code, Jupyter, and a terminal. It supports custom images, persistent storage, and remote debugging, making it suitable for AI model development. | DevPod |
| Elastic policy for resident resource pools | New | GPU resident resource pools now support request-based elasticity, scheduled elasticity, and metric-based scaling to improve resource utilization. | Instance scaling limits and elastic policies |
| Ulanqab compliance cloud launch | New | Dedicated cloud and general industry cloud are now available in the Ulanqab region. | Available regions and Service endpoints |
| Light hibernation billing for delayed release on-demand instances | New | Both CPU and GPU on-demand instances support light hibernation billing in delayed release mode. GPU on-demand instances do not support light hibernation for WebSocket connections. | Configure delayed release for on-demand instances |
October 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| Delayed release for on-demand instances | New | Keep instances alive after a request completes to finish background tasks such as log uploads or data synchronization. After the delayed release window, the system automatically keeps the on-demand instance active, switches it to an idle state, or destroys it based on vCPU utilization. | Configure delayed release for on-demand instances |
| Dynamic NAS mounting | New | Mount each user session in a function instance to a dedicated subdirectory in File Storage NAS (NAS), providing secure isolation for persistent data in multi-tenant environments. | Dynamically mount File Storage NAS |
| New GPU types: Hopper.2 and Blackwell.1 | New | Added support for two new GPU instance types: Hopper.2 and Blackwell.1. | Instance types and specifications |
September 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| gRPC support for HTTP triggers | New | Trigger gRPC services directly through HTTP triggers. The function acts as a gRPC server and processes both streaming and non-streaming client requests, while benefiting from serverless auto scaling. | Configure an HTTP trigger and use gRPC to invoke functions and gRPC protocol support for custom domain names |
| Model Context Protocol (MCP) Streamable HTTP affinity | New | Route requests from the same MCP session to the function instance that initialized the session. FC parses the Mcp-Session-Id field in the HTTP response header and uses it to associate subsequent requests with the correct instance. | Configure MCP Streamable HTTP affinity |
| Session lifecycle management | New | Manage HeaderField and Cookie affinity sessions as explicit resources with full lifecycle APIs: create, query, update, and delete. | Session lifecycle management |
August 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| Tracing Analysis | New | FC integrates with Alibaba Cloud Managed Service for OpenTelemetry (based on the W3C OpenTelemetry standard) to identify and diagnose performance bottlenecks in distributed applications. | Configure Tracing Analysis |
Minimum instances | Optimization | To better meet the needs of stateful and background services, the snapshot feature is upgraded to the minimum instances feature. Set the minimum number of instances to a value greater than 0 to resolve cold start issues for on-demand instances. This also lets you lock elastic resources in advance to ensure guaranteed delivery. Features like session affinity scheduling and persistent connections ensure stability for highly interactive real-time applications. | |
| Basic authentication | New | Require clients to include Base64-encoded username and password credentials in requests to a custom domain name or HTTP trigger. Access is granted only after the credentials are verified. | Configure Basic authentication for a custom domain name and Configure Basic authentication for an HTTP trigger |
July 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| Session affinity | New | Route requests from the same client to the same function instance for a configurable period, maintaining session continuity for stateful services. | Configure MCP Streamable HTTP affinity, Configure MCP SSE affinity, Configure HeaderField affinity, and Configure Cookie affinity |
| Instance isolation | New | Restrict an instance to processing only a single request, or all requests within a single session. The instance is released when the request or session ends, providing an isolated sandbox runtime environment. | Overview of instance isolation and Best practices for configuring instance isolation |
Snapshot feature | New | Use snapshots to resolve cold start issues when on-demand instances process their first request. You can lock elastic resources such as CPU and GPU in advance to ensure guaranteed delivery. This feature also addresses issues like session affinity and persistent connection keepalive for stateful applications. | |
| Resource group support | New | Organize FC resources into resource groups by department, project, or environment. Combined with Resource Access Management (RAM), resource groups enable resource isolation and fine-grained permission control within a single Alibaba Cloud account. | Configure resource groups |
| Resident resource pools (subscription) | New | Purchase a resident resource pool to reserve compute capacity of a specific type in advance. Assign resident instances from the pool to functions as needed for predictable performance and fixed costs. | Resident resource pools (subscription), Configure resident instances, and Select an instance type |
May 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| Command-type Initializer hook | New | Configure a command in the Initializer hook to send an HTTP request to a function after the instance starts — for example, to prefetch a model in an AI inference application. The Initializer hook now supports a command type in addition to the code invocation type. Currently supported on GPU functions only. | |
| Bearer authentication | New | Configure allowed tokens in the FC console. Clients must include a valid Bearer token in the Authorization header. Requests are granted access only if the token matches the configured value on the trigger. | Enable Bearer authentication for an HTTP trigger and Configure Bearer authentication for a custom domain name |
April 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| Enhanced function management | Optimization | The function list in the FC console now includes tag and specification columns with sorting. You can also customize columns and use flexible paging to display functions. Search functions by name, description, runtime, specification, or tags — and combine criteria. Batch-edit tags across multiple functions. | Configure tags |
| Function Compute 3.0 support for OSS object access points | New | Automatically trigger a function when an OSS GetObject event occurs on an access point. The function processes the request and returns the result to the application, enabling automated data processing workflows. | Trigger a function with a GetObject request using an Object FC Access Point |
| Function Compute 3.0 support for ZIP file decompression | New | Configure decompression rules on an OSS bucket. When a matching ZIP file is uploaded, FC automatically triggers a function to decompress the file and write the contents to the target directory. | Configure a ZIP file decompression trigger |
March 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| Application Load Balancer (ALB) trigger | New | Use an ALB instance as an event source. Add a server group of the FC type to the ALB to forward incoming requests to Function Compute and invoke a function. | ALB trigger |
February 2025
| Feature | Change type | Description | References |
|---|---|---|---|
| gRPC support for custom domain names | New | Bind a custom domain name to a function and use a gRPC client to send gRPC requests. The function acts as a gRPC server and returns results to the client. | gRPC protocol support for custom domain names |
该文章对您有帮助吗？