Virtual Node Architecture for Serverless Elastic Pod Scheduling - Container Compute Service

ACS Virtual Node connects your Kubernetes cluster to the serverless compute resources of Alibaba Cloud Container Compute Service (ACS), letting you schedule pods onto on-demand infrastructure alongside—or instead of—traditional nodes.

ACS Virtual Node is built on the open-source Virtual Kubelet project, extended with ACS provider support and optimized for serverless container workloads.

Key capabilities

Scale without node limits

Pods scheduled to a virtual node run as ACS serverless containers. Cluster elasticity is no longer bounded by node capacity: spin up thousands of pods on demand without provisioning or managing underlying VMs.

Pay only for what you run

ACS Pods use pay-as-you-go billing. You pay for actual execution time, not idle node capacity between workload bursts.

Eliminate node operations

Serverless containers remove the operational overhead of OS patching, node scaling, and capacity planning.

Use cases

ACS Virtual Node works well for workloads that are bursty, short-lived, or isolated:

Use case	Description
Batch jobs	Run jobs to completion without maintaining long-running node fleets.
CI/CD pipelines	Spin up ephemeral build and test environments on demand.
Big data processing	Scale Apache Spark jobs elastically without pre-provisioning large clusters.
Online application scaling	Handle traffic spikes by bursting beyond fixed node capacity.

Usage

For ACS Pod features and configuration details, see ACS Pod instance overview.

Release history

March 2026

Version	Date	Description	Impact
v2.16.5	2026-03-20	Optimized Pod update efficiency at scale.	This update does not affect your workloads.

February 2026

Version	Date	Description	Impact
v2.16.4	2026-02-26	Optimized pod creation efficiency in large-scale scenarios. Optimized the efficiency of binding pods to nodes in large-scale scenarios.	This update does not affect your workloads.
v2.16.3	2026-02-05	Optimizations: Optimized the startup efficiency of the component in large-scale clusters. Optimized pod deletion efficiency in large-scale clusters.	This update does not affect your workloads.

January 2026

Version	Date	Description	Impact
v2.16.2	2026-01-28	Optimizations: Optimized the retry mechanism for pod creation requests on virtual nodes when the quota is exceeded. Bug fixes: Fixed an issue where pods on virtual nodes would occasionally get stuck in the Pending state due to the presence of unschedulable reserved nodes.	This update does not affect your workloads.
v2.16.1	2026-01-27	Optimizations: Optimized the internal HTTP connection pool management mechanism of the component. Optimized the pod creation performance of version v2.16.0 in large-scale scenarios and reduced startup time. Bug fixes: Fixed an issue where pods would occasionally get stuck in the Pending state in extreme cases.	This update does not affect your workloads.
v2.16.0	2026-01-06	New features: Enabled the Pod Capacity feature for virtual nodes by default. Added support for using the PodNetworking feature with ACS computing power. Added support for referencing resources across namespaces using `NodePublishSecretRef` with ACS computing power. Added support for enabling the CSINode feature for reserved nodes. Added support for sending event notifications during the self-healing process of reserved nodes. Added support for displaying the `InterruptionHardwareFault` condition on pods of ACS GPU-accelerated instances and reserved nodes. Optimizations: Optimized the startup speed of the component in large-scale scenarios. Improved the success rate of cleaning up instances in extreme cases. Bug fixes: Fixed an issue where reserved nodes might enter the `NotReady` state in specific scenarios. Fixed an issue where pod conditions might be unexpectedly overwritten in extreme cases. Fixed an authorization issue that might occur when using SidecarSet injection with ACS computing power. Fixed an issue where the `Instance.ToBeReleased` condition was not correctly displayed on ACS computing power instances.	This update does not affect your workloads.

December 2025

Version	Date	Description	Impact
v2.15.8	2025-12-08	Optimizations: Fixed an issue where the local Informer cache expired. Optimized the processing speed of pods in the desired state. Bug fixes: Fixed an issue where permissions for SidecarSet pods were revoked. Fixed an issue where conditions were overwritten in extreme cases.	This update does not affect your workloads.

November 2025

Version	Date	Description	Impact
v2.15.7	2025-11-26	Optimizations: Optimized the rate limiting and retry mechanism. Bug fixes: Fixed an issue where the IP addresses of reserved nodes were not updated.	This update does not affect your workloads.
v2.15.6	2025-11-05	Optimizations: Optimized the resource normalization logic for ACS instances. Optimized the error message that appears when a pod on a reserved node uses an unsupported `ComputeClass`. Bug fixes: Fixed an issue where the local Informer cache might not expire promptly. Fixed an issue where a pod's annotation might become inconsistent with its `Spec` definition after an in-place resource change (InPlace VPA).	This update does not affect your workloads.

October 2025

Version	Date	Description	Impact
v2.15.5	2025-10-20	Bug fixes: Fixed an issue where the IP addresses of reserved nodes in ACS clusters were missing.	This update does not affect your workloads.

September 2025

Version	Date	Description	Impact
v2.15.4	2025-09-28	Improvements: Improved the error message for DaemonSet pods scheduled to a virtual node. Improved the recovery of GPU pods from abnormal states in an ACK cluster. Bug fixes: Resolved an issue where the status of an ACS Pod would occasionally not update. Fixed an issue where the IP address of a reserved node was missing in an ACK cluster. Resolved an occasional crash when using CNFS volumes.	This update does not affect your workloads.
v2.15.3	2025-09-28	Bug fixes: Resolved an issue where pods remained in the `Pending` state for an extended period in rare scenarios. Fixed a bug that prevented pods from reaching a terminal state (such as `Succeeded` or `Failed`) in rare scenarios.	This update does not affect your workloads.
v2.15.2	2025-09-22	Bug fixes: Fixed an issue in v2.15.0 where ACS resource normalization did not take effect. Corrected the inaccurate `KubeletVersion` field for reserved nodes. Resolved an issue with expired local cache data caused by API server push delays.	This update does not affect your workloads.
v2.15.1	2025-09-09	Bug fixes: Resolved an issue where pod creation failed when specifying a vSwitch by using an annotation.	This update does not affect your workloads.

August 2025

Version	Date	Description	Impact
v2.15.0	2025-08-19	New features: Added support for the reserved node feature. Added automatic protocol conversion from ECI to ACS. Improvements: Reduced memory consumption by approximately 40%. Improved the creation success rate of ACS Pods. Adjusted the ephemeral storage capacity (`ephemeral-storage`) of the virtual node to 60,000 TiB to prevent scheduling failures. Bug fixes: Resolved an issue where resizing a cloud disk attached to an ACS GPU Pod failed. Fixed an issue where the in-place container restart process failed. Resolved an issue where the `PodNetworking` configuration did not take effect in an ACK cluster.	This update does not affect your workloads.
v2.12.0-acs.16	2025-08-07	Resolved an issue where pods would get stuck in the Pending state under abnormal conditions. Added support for reserved node status reporting. Optimized reporting of the `Interruption.GPUShareReclaim` condition.	This update does not affect your workloads.

May 2025

Version	Date	Description	Impact
v2.12.0-acs.14	2025-05-12	Fixed unexpected update behavior for a reserved node. Resolved a parsing exception caused by an invalid acs-profile configuration format. Added support for exposing the `Interruption.GPUShareReclaim` condition.	This update does not affect your workloads.

March 2025

Version	Date	Description	Impact
v2.12.0-acs.13	2025-03-20	Enabled inventory-aware scheduling for GPU pods.	This update does not affect your workloads.

February 2025

Version	Date	Description	Impact
v2.12.0-acs.12	2025-02-13	Adjusted the ephemeral storage capacity on the virtual node to 60,000 TiB. Resolved an issue where retries were not attempted after a ServiceUnavailable error occurred in advanced scheduling scenarios. Optimized the interaction with the csi-provisioner component when a PersistentVolumeClaim (PVC) uses the WaitForFirstConsumer binding mode.	This update does not affect your workloads.

January 2025

Version	Date	Description	Impact
v2.12.0-acs.11	2025-01-21	Enhanced advanced scheduling capabilities.	This update does not affect your workloads.
v2.12.0-acs.10	2025-01-17	Added support for exposing metrics from a reserved node. Optimized the retry mechanism. Other enhancements.	This update does not affect your workloads.

November 2024

Version	Date	Description	Impact
v2.12.0-acs.9	2024-11-27	Added support for resizing cloud disks for GPU pods. Enabled inventory awareness for general-purpose and performance-optimized pods. Other enhancements.	This update does not affect your workloads.

September 2024

Version	Date	Description	Impact
v2.12.0-acs.8	2024-09-13	Upgraded the Kubernetes API dependency to v1.30. Added the `container_sockets` and `container_processes` Prometheus metrics. Added support for network metric collection by using KubeSkoop. Added support for attaching a cloud disk to GPU pods. Enabled CPU hot-swapping for general-purpose pods.	This update does not affect your workloads.

August 2024

Version	Date	Description	Impact
v2.12.0-acs.7	2024-08-21	Resolved an issue where an associated event was missing a UID.	This update does not affect your workloads.
v2.12.0-acs.6	2024-08-02	Enabled resizing for cloud disks attached to general-purpose and performance-optimized pods.	This update does not affect your workloads.

July 2024

Version	Date	Description	Impact
v2.12.0-acs.5	2024-07-25	Replaced Role and RoleBinding authentication with Pod identity-based authentication.	This update does not affect your workloads.

June 2024

Version	Date	Description	Impact
v2.12.0-acs.4	2024-06-21	Integrated with the scheduler to support Kubernetes-native scheduling semantics. Added support for creating pods by using resources from a capacity reservation.	This update does not affect your workloads.