This page covers ACK feature releases in 2023, organized by month. For the latest release notes, see Release notes.
December 2023
| Product | Feature | Description | Region | References |
|---|---|---|---|---|
|
Container Service for Kubernetes |
Support for AI model inference protection based on confidential virtual machines |
ACK now supports adding ECS instances that use Intel® Trusted Domain Extension (Intel® TDX) technology to a TDX confidential virtual machine compute node pool. This gives the cluster TDX confidential computing capabilities to enable trusted inference and fine-tuning of AI models, which ensures the confidentiality and integrity of model data. By combining PyTorch with the Intel® AMX instruction set, you can achieve image generation in seconds on a 32 core instance. |
All |
|
|
Launched in China (Wuhan) Local Region |
The service is now available in three local regions: Nanjing, Fuzhou, and Wuhan. Local regions use the same Apsara architecture as central regions and feature a small footprint, consistency, low latency, and low cost. They are designed to support scenarios such as local data processing, low-latency lift-and-shift migration of IT center architectures, and real-time data interaction. |
China (Wuhan) Local Region |
||
| Cloud-native AI suite | ack-ai-installer 1.8.7 | Adds Multi-Process Service (MPS)-based GPU sharing. The ack-ai-installer component integrates GPU sharing and topology-aware GPU scheduling capabilities. cGPU is updated to 1.5.5. | All | ack-ai-installer |
| Cloud-native AI suite | ack-fluid 1.0.6 | Supports configuring resource requests and limits, creating custom multi-level affinity-based scheduling policies, and configuring JindoRuntime to accelerate access to persistent volumes (PVs) and host directories. | All | ack-fluid |
| Cloud-native AI suite | ack-kube-queue 0.3.3 | Supports configuring blocking queues using environment variables, with the blocking queue mode applied globally across all queues. | All | ack-kube-queue |
| ACK Edge | AdvancedRollingUpdate and OTA update models for DaemonSets | Introduces AdvancedRollingUpdate and over-the-air (OTA) update models for DaemonSets. Standard rolling updates can stall when edge nodes enter NotReady state after a network disconnection. These models address that issue and support OTA updates for edge devices such as green vehicles. | All | DaemonSet update models |
| ACK Edge | Region-specific workload customization | Adds the YurtAppOverrider CustomResourceDefinition (CRD), which works with YurtAppSet to deploy different replica counts or image versions for the same workload across regions. | All | — |
| ACK Edge | Raven component (replaces edge-tunnel) | Replaces edge-tunnel with Raven for cloud-edge communication in ACK Edge clusters. Raven provides proxy mode for HTTP host network communication and tunnel mode for cross-domain container network communication between node pools in different regions. | All | Cross-region O&M communication component Raven |
| ACK One | Quick start for registered clusters | Adds a quick start guide for getting started with registered clusters. | All | Getting started with registered clusters |
| ACK One | Eventing for workflow clusters | ACK One workflow clusters now support eventing, enabling event-driven automation. Supported event sources include Object Storage Service (OSS), Simple Message Queue (formerly MNS), Git repositories, and EventBridge. | All | Enable eventing |
November 2023
| Product | Feature | Description | Region | References |
|---|---|---|---|---|
| Container Service for Kubernetes | Automated O&M for managed node pools | Auto node recovery, auto kubelet update, and auto CVE patching can now be enabled when creating or modifying a managed node pool. | All | Create and manage a node pool |
| ACK One | Zone-disaster recovery with multi-cluster gateways | The ACK One GitOps multi-cluster application distribution feature now works with ACK One multi-cluster gateways to support zone-disaster recovery. | All | — |
October 2023
| Product | Feature | Description | Region | References |
|---|---|---|---|---|
| Container Service for Kubernetes | Sidecar acceleration in service mesh scenarios | Accelerates TCP communication between sidecar containers on the same node, and between regular containers and sidecar containers in the same pod, using the eBPF-based sidecar acceleration component. | All | Sidecar acceleration using eBPF |
| Container Service for Kubernetes | OS kernel-level container monitoring | Provides OS kernel-level monitoring to make the container engine layer more transparent, helping you troubleshoot container service failures and migrate containerized applications. | All | Use SysOM to locate container memory issues |
| Container Service for Kubernetes | Individual cluster inspection | Supports on-demand cluster inspection by calling the RunClusterCheck API operation. Run update prechecks at any time and create update plans based on the results. | All | — |
| Container Service for Kubernetes | ARM64 support for migrate-controller | The migrate-controller backup center component now supports the ARM64 architecture. The component is based on open-source Velero and integrates with Alibaba Cloud snapshot technologies and Cloud Backup to back up, restore, and migrate data and application templates. | All | migrate-controller |
| ACK One | Multi-cluster gateways | ACK One uses a fully managed MSE Ingress controller to manage MSE cloud-native gateways as multi-cluster gateways. Supports active zone-redundancy, multi-cluster load balancing, and header-based routing for north-south traffic. | All | Overview of multi-cluster gateways |
September 2023
| Product | Feature | Description | Region | References |
|---|---|---|---|---|
| ACK | ack-advanced-audit 0.3.0 | Improves performance for in-container operation auditing and adds support for the latest Alibaba Cloud Linux kernel versions. | All | ack-advanced-audit |
| Cloud-native AI suite | Push notebook images to Container Registry | Use ack-ai-dev-console with commit-agent and ack-commit-ctl to save notebooks running on Elastic Compute Service (ECS) instances as images and push them to Container Registry. Notebooks can then be recreated from those image addresses. Private repositories require credential-based image pulling. | All | Save a notebook as an image |
| Cloud-native AI suite | KServe inference with Arena and Service Mesh | Deploy KServe inference services in ACK clusters using the Arena CLI together with Alibaba Cloud Service Mesh (ASM). Enable KServe on ASM in the ASM console to install the component, then submit inference services through Arena. KServe supports auto scaling, blue-green deployment, canary releases, and scale-to-zero. | All | — |
| Cloud-native AI suite | ack-fluid 1.0.4 | Supports launching the FUSE sidecar plug-in in asynchronous mode, dynamically updating JuiceFSRuntime configurations, and is now compatible with ACK Lingjun clusters. | All | ack-fluid |
| Cloud-native AI suite | ack-kube-queue 0.3.1 | Supports retrieving job sequence information from queues, adds --qps and --burst settings for the batchjob-extension component, and fixes an error that occasionally occurred when deleting QueueUnit. |
All | ack-kube-queue |
| ACK One | ACK One GitOps best practice | Publishes a best practice guide for building CI/CD pipelines across Dev, Staging, and Production clusters using ACK One GitOps and Container Registry. Pipelines automatically sync image changes and deploy updates after code is committed to a Git repository. | All | — |
| ACK One | On-premises storage access with Fluid | Fluid's ThinRuntime CRD can now describe custom storage systems and interface them with Fluid, enabling data access from on-premises storage in cloud-native scenarios. | All | Use Fluid to access on-premises storage |
| ACK One | OSS access acceleration with Fluid | Fluid accelerates access to OSS data using JindoRuntime, the C++-based execution engine of JindoFS developed by the Alibaba Cloud E-MapReduce (EMR) team. JindoRuntime provides dataset management and caching for OSS-backed datasets. | All | Use Fluid to accelerate access to OSS objects |
| ACK Edge | NodePort listening based on node pools | NodePort Services can now listen only on nodes within a specified node pool, preventing port conflicts and enabling service isolation in edge computing scenarios. | All | Configure NodePort listening based on node pools |
| ACK Lingjun | Available on Alibaba Cloud International site | ACK Lingjun clusters are now available on the Alibaba Cloud International site. The ack-fluid component can be deployed from the App Catalog page in the ACK console. | All | What is ACK Lingjun? |
August 2023
| Product | Feature | Description | Region | References |
|---|---|---|---|---|
| ACK | ACK backup center improvements | Supports fuzzy search for backup tasks, sorting by creation time, and cloning real-time backup tasks and backup plans. | All | Back up and restore applications |
| ACK Serverless | SidecarSet controller in ack-virtual-node | Adds the SidecarSet controller to ack-virtual-node, enabling sidecar injection into pods on virtual nodes. This allows workloads to use DaemonSet-equivalent features such as log collection and monitoring. | All | — |
| Cloud-native AI suite | Dynamic MIG partitioning for GPU scheduling | Supports multi-instance GPU (MIG) and dynamic partitioning on GPU nodes with the required label after installing ack-ai-installer. Each container can request one GPU instance, and each node reports the maximum number of supported GPU instances. | All | Labels for GPU-accelerated node scheduling and methods for changing label values |
| Cloud-native AI suite | Notebook images in AI Developer Console | Adds commit-agent and ack-commit-ctl to AI Developer Console (ack-ai-dev-console), allowing notebooks running on ECS instances to be saved as images and later restored from those images. | All | Create and use a Jupyter notebook |
| Cloud-native AI suite | ack-fluid 1.0.3 | Supports privileged FUSE sidecar containers. Fixes an out-of-memory (OOM) issue in the JuiceFSRuntime controller, issues with the FUSE auto recovery feature, and residual resources after runtime updates. | All | ack-fluid |
| Cloud-native AI suite | ack-kube-queue 0.2.0 | Supports enabling blocking queues for a single queue, re-adding TensorFlow and PyTorch jobs to the queue on timeout, adding Argo workflows to a queue, and limiting the number of jobs that can leave a queue at a time. Improves dequeue failure logging. | All | ack-kube-queue |
| ACK One | SLB management with CCM in registered clusters | Install the cloud controller manager (CCM) in registered clusters to manage Server Load Balancer (SLB) instances, including Classic Load Balancer (CLB). | All | Use CCM in registered clusters |
| ACK One | Custom DNS servers for registered clusters | Specify self-managed DNS servers for elastic container instances in registered clusters. | All | Specify DNS servers for registered clusters |
| ACK One | Public access to Argo CD | Enable public access to Argo CD from the ACK One console to deliver and access applications through the Argo CD interface. | All | Enable public access to Argo CD |
July 2023
| Feature | Description | Region | References |
|---|---|---|---|
| ACK Standard clusters renamed to ACK Basic clusters | ACK Standard clusters are renamed to ACK Basic clusters. This affects ACK managed clusters, ACK Serverless clusters, and ACK Edge clusters. The corresponding API parameters are unchanged. ACK Basic clusters are for personal use and testing. Use ACK Pro clusters in production environments. | All | Hot migration from ACK Basic clusters to ACK Pro clusters |
| ALB Ingress controller v2.9.0-aliyun.1 released | ALB Ingress controller v2.9.0-aliyun.1 is now available. | All | ALB Ingress controller |
| Terway 1.5.5 released | Terway 1.5.5 adds support for elastic network interface (ENI) filters. | All | Configure an ENI filter |
| GPU-accelerated node diagnostics | Collect basic GPU metrics and troubleshoot GPU issues at the GPU level in ACK clusters. | All | Diagnose GPU-accelerated nodes |
| IP-aware and topology-aware pod scheduling | IP-aware scheduling: Blacklists a node for five minutes when it has insufficient idle IP addresses, preventing scheduling failures under high pod load. Applies only to ACK Pro clusters using Terway. Topology-aware scheduling: Schedules a group of pods with the same topology constraints to different topology domains, reducing response latency and improving availability. Elastic quota resource info: Query resource requests and usage per elastic quota in an ElasticQuotaTree through Capacity Scheduling. | All | — |
| ACK backup center improvements | Supports excluding specific namespaces when selecting backup scope, and creating and copying backup plans. | All | Backup center |
| Serverless Kubernetes renamed to ACK Serverless | ASK (Serverless Kubernetes) is renamed to ACK Serverless, and ASK clusters are renamed to ACK Serverless clusters. ACK Serverless clusters are billed by uptime in seconds and support auto scaling within seconds, with no cluster or node maintenance required. | All | ASK renamed as ACK Serverless |
| KServe in ACK Serverless clusters | Deploy KServe in ACK Serverless clusters with a few clicks to serve trained models. Supported runtimes include TFServing, TorchServe, and Triton. KServe provides out-of-the-box model services and supports custom runtimes. | All | Deploy the KServe component and Quickly deploy an inference service based on KServe |
| Fully managed core components in ACK Serverless Pro clusters | The Kube scheduler, CCM, Kube controller manager, Kube API server, and CoreDNS are fully managed in ACK Serverless Pro clusters, consuming none of your cluster resources. | All | What is ACK Serverless? |
| Preemptible instance creation for large-scale jobs | ACK Serverless supports creating instances with up to one million CPU cores within one hour to handle large concurrent workloads such as Spark jobs in big data scenarios where compute and storage are decoupled. | All | Use ACK Serverless to create Spark jobs |
| ack-kube-queue 0.1.12 released | Adds per-queue blocking policy configuration via environment variables in the Kube queue controller. Adds jobrunningtimeout and jobbackofftime parameters for dequeue timeout control: when a job fails to start within jobrunningtimeout, it enters backoff state and is re-added to the queue after jobbackofftime. Supports only TensorFlow and PyTorch jobs. |
All | — |
| ack-fluid 1.0.2 released | Adds security hardening for Fluid control plane components, local disk caching for elastic container instances, and scheduled data preloading via DataLoad. | All | — |
| Registered cluster creation with onectl | Use onectl to create registered clusters, connect them to external Kubernetes clusters, configure permissions, and install components. | All | Use onectl to create a registered cluster and Use onectl to manage registered clusters |
| Serverless ECI resources for external Kubernetes clusters | Registered clusters let external Kubernetes clusters in data centers access CPU and GPU resources from serverless Elastic Container Instance (ECI). | All | Use registered clusters to allow external Kubernetes clusters to use CPU and GPU resources provided by serverless elastic container instances |
June 2023
| Feature | Description | Region | References |
|---|---|---|---|
| ACK clusters support Kubernetes 1.26 upgrades | ACK clusters can now be updated to Kubernetes 1.26. | All | Kubernetes 1.26 |
| Network diagnostics in Container Intelligence Service | Network diagnostics based on Skoop are now available in Container Intelligence Service. | All | Network diagnostics |
| ACK backup center enhancements | Adds backup plan editing and advanced configuration display for backup tasks and snapshots. | All | Backup center |
| ACK scheduler update | Adds topology-aware GPU scheduling for Kubernetes 1.26 clusters, updates custom elastic resource priorities, and supports setting the maximum number of pods per resource level. | All | kube-scheduler |
| Cloud-native AI suite in ACK Serverless Pro clusters | Install and activate the cloud-native AI suite in ACK Serverless Pro clusters to run AI, machine learning, and big data workloads. Arena handles job submission for mainstream AI frameworks. Fluid provides data elasticity and preloading for serverless scenarios. | All | Deploy the cloud-native AI component set |
| Accelerate access to PVs and hostPath volumes with Fluid | JindoRuntime caches data from Kubernetes PVs and hostPath volumes to accelerate access. In hybrid cloud environments, use PVs to mount self-managed storage systems such as CephFS, or hostPath volumes to mount other self-managed storage. | All | Use JindoRuntime to accelerate access to PVs and Accelerate access to hostPath volumes |
| DeepSpeed distributed training with Arena | Submit and manage DeepSpeed distributed training jobs using Arena. Visualize jobs with TensorBoard. DeepSpeed is an open-source deep learning optimization suite that provides distributed training and model optimization. | All | DeepSpeed distributed training |
| ARM-based node support in ack-kube-queue | ack-kube-queue now supports ARM-based nodes, enabling queue management and priority-based scheduling for AI, machine learning, and batch computing workloads on ARM architecture. | All | Use ack-kube-queue to manage AI and machine learning workloads |
| ACK One: Application migration to ACK One GitOps | Migrate applications from Application Center to ACK One GitOps, which provides a dedicated console with Alibaba Cloud account single sign-on (SSO) for multi-cluster application delivery. | All | Migrate applications to ACK One GitOps |
| ACK One: Policy management for registered clusters | Registered clusters now support a policy governance console based on Open Policy Agent (OPA) and the gatekeeper admission controller. | All | Configure ACK pod security policies in registered clusters |
| ACK One: Multi-cluster GitOps best practice | Publishes a best practice guide for multi-cluster application release, update, and rollback using ACK One GitOps, including continuous integration (CI) pipeline integration and access control. | All | Use ACK One GitOps to deploy applications |
May 2023
| Feature | Description | Region | References |
|---|---|---|---|
| Kubernetes 1.26 support | ACK now supports creating clusters that run Kubernetes 1.26. | All | Kubernetes 1.26 |
| csi-compatible-controller | Use FlexVolume and CSI in the same ACK cluster, and migrate volumes from FlexVolume to CSI without changing application configurations. | All | csi-compatible-controller and Use csi-compatible-controller to migrate from FlexVolume to CSI |
| Specified ECS instance types for elastic container instances (ACK Serverless) | Create elastic container instances from specified ECS instance types. These instances support multiple processors and reduce instance cost by 40% compared with previous-generation instances. Purchase savings plans through annual subscriptions for an additional 10% reduction on unexpected traffic. | All | Create an elastic container instance from a specified ECS instance type |
| Cost insights for ACK Serverless | Analyze elastic resource costs by cluster, namespace, and application. Supports fine-grained cost prediction and root-cause analysis to help optimize resource allocation and billing. | All | Cost Insights |
| Blocking queues and strict priority scheduling in ack-kube-queue | Blocking queues: the queue processes one job at a time until it exits before moving to the next. Strict priority scheduling: resources go to higher-priority jobs first, then to earlier-submitted jobs, ensuring backoff jobs are eventually scheduled. | All | Use ack-kube-queue to manage AI and machine learning workloads |
| VS Code notebooks in cloud-native AI suite | Create VS Code notebooks to develop and debug machine learning algorithms, then submit them to ACK clusters for training. Supports JupyterLab, custom Jupyter notebook images, and custom VS Code notebook images, integrated with ACK's permission management system. | All | Create and use a Jupyter notebook |
| ACK One GitOps enable/disable | Enable or disable the GitOps console from the master instance page in the ACK One console, with a direct link to the GitOps console. | All | Log on to the GitOps system |
| ACK One GitOps control plane and audit logs | View GitOps control plane logs and audit logs from the master instance page in the ACK One console. | All | Enable the collection of the control plane logs and audit logs of GitOps |
April 2023
| Feature | Description | Region | References |
|---|---|---|---|
| Delete Simple Log Service (SLS) projects with clusters | When deleting a cluster, optionally delete the associated Simple Log Service (SLS) project at the same time. Important
An SLS project includes multiple Logstores. Deleting the project permanently deletes all Logstore data. |
All | Delete a cluster |
| ACK scheduler 4.1 | Elastic scheduling now supports scheduling pods to ARM-based nodes using elastic resources. To prevent resource leaks: cluster-autoscaler is blocked from creating nodes by default when the elastic quota is exhausted in capacity scheduling scenarios, or when gang scheduling requirements are unmet. | All | kube-scheduler |
| Fluid control plane monitoring | Install the Fluid monitoring component on Managed Service for Prometheus instances from the ACK console, then use out-of-the-box dashboards to monitor Fluid control plane components. | All | Enable Managed Service for Prometheus for the Fluid component |
| Multi-tenant access control for ACK One GitOps | Configure role-based access control for ACK One GitOps to manage permissions for different roles and reduce security risks. | All | Configure multi-tenant access control for ACK One GitOps |
March 2023
| Feature | Description | Region | References |
|---|---|---|---|
| vSwitch configuration in Terway | Add or remove vSwitches by modifying the terway-eniip component on the Add-ons page in the ACK console. | All | Modify the pod vSwitches |
| NAS and CPFS client I/O troubleshooting with CNFS dashboards | Use Container Network File System (CNFS) dashboards to troubleshoot client I/O issues such as excessive I/O frequency or high bandwidth usage, and identify the pods causing them. | All | Use the observabilities of CNFS to identify the issues caused by I/O operations on the NAS or CPFS client |
| ACK One RAM user permission management | Authorize Resource Access Management (RAM) users to access and manage ACK One master instances using administrator accounts. | All | Grant permissions to a RAM user |
| ACK One GitOps Image Updater | ACK One GitOps Image Updater monitors Container Registry for image updates and automatically triggers application updates when new image versions are pushed. Enables GitOps pipelines integrated with third-party CI systems. | All | — |
| Cloud-native AI suite pricing update | Tiered pricing takes effect at 00:00:00 (UTC+8) on April 1, 2023. The 50% discount ends at that time. | All | \[Product changes\] Cloud-native AI suite metering method and pricing solution updates and Billing of the cloud-native AI suite |
| ack-fluid 0.9.7 released | Fixes known issues and adds security reinforcement. | All | \[Component updates\] Update ack-fluid |
| Advanced features for gang scheduling | Adds GangGroup support, which groups multiple gangs so all-or-nothing scheduling applies across all roles in a job. Each role's min-available requirement must be met before the job runs. Gang scheduling retries can be controlled using a matchpolicy. |
All | Use gang scheduling |
February 2023
| Feature | Description | Region | References |
|---|---|---|---|
| Custom kubelet parameters | Customize kubelet configurations per node pool. Changes take effect immediately on existing nodes in batches and are automatically applied to newly added nodes. | All | Customize the kubelet parameters of a node pool |
| Service diagnostics in AIOps | Troubleshoot Classic Load Balancer (CLB) issues related to billing methods, certificates, CLB instance reuse, CLB quotas, and anomaly events. | All | Service diagnostics |
| Cloud disk type changes | Change the disk type mounted to an application — for example, upgrade from standard SSD to Enhanced SSD (ESSD) for higher IOPS. | All | Change the category of a cloud disk |
| cloud_auto disk type for dynamically provisioned volumes | Dynamically provisioned volumes now support the cloud_auto disk type, in addition to cloud_efficiency, cloud_ssd, cloud_essd, and available. | All | Use a dynamically provisioned disk volume |
| Alibaba Cloud accounts for Argo CD single sign-on | After logging in to the ACK One console with an Alibaba Cloud account, access the Argo CD UI or Argo CD CLI via single sign-on (SSO) without entering credentials. | All | Log on to the GitOps system |
Historical releases
For ACK release notes before 2023, see Historical release notes (before 2025).
该文章对您有帮助吗？