All Products
Search

This Product

    Container Compute Service:ACS Pod instance overview

    Document Center

    Container Compute Service:ACS Pod instance overview

    Last Updated:Mar 26, 2026

    Container Compute Service (ACS) runs pods as the basic unit of deployment. Each ACS pod is a fully isolated, serverless container runtime backed by lightweight sandboxed container technology—no node management required. This topic covers compute types, resource specifications, core capabilities, and Kubernetes limitations for ACS pods.

    Compute types

    ACS supports four compute types. Use the alibabacloud.com/compute-class label to specify the compute type for a pod.

    Compute type Label Best for
    General-purpose (default) general-purpose Stateless microservices, Java web applications, and compute-intensive tasks
    Compute-optimized performance CPU-based AI/ML training and inference, and High Performance Computing (HPC) batch processing
    GPU gpu Single-GPU and multi-GPU inference, and GPU parallel computing
    GPU-HPN (High-Performance Network GPU) gpu-hpn GPU distributed training, distributed inference, and GPU high-performance computing

    The following YAML examples show how to set the compute type for a Deployment.

    General-purpose

    apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
        alibabacloud.com/compute-class: general-purpose
    spec:
      containers:
      - name: nginx
        image: registry.cn-hangzhou.aliyuncs.com/acs-sample/nginx:latest

    GPU

    apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
        # Specify the compute type as gpu
        alibabacloud.com/compute-class: "gpu"
        # Specify the GPU model series. For example: T4
        alibabacloud.com/gpu-model-series: "example-model"
    spec:
      containers:
      - name: nginx
        image: registry.cn-hangzhou.aliyuncs.com/acs-sample/nginx:latest
        resources:
          limits:
            cpu: 4
            memory: "8Gi"
            nvidia.com/gpu: "1"
          requests:
            cpu: 4
            memory: "8Gi"
            nvidia.com/gpu: "1"

    For supported GPU models and specifications, see Accelerated compute types.

    GPU-HPN

    apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
        # Specify the compute type as gpu-hpn
        alibabacloud.com/compute-class: "gpu-hpn"
    spec:
      containers:
      - name: nginx
        image: registry.cn-hangzhou.aliyuncs.com/acs-sample/nginx:latest
        resources:
          limits:
            cpu: 4
            memory: "8Gi"
            nvidia.com/gpu: "1"
          requests:
            cpu: 4
            memory: "8Gi"
            nvidia.com/gpu: "1"
    To use GPU-HPN instances, first create a GPU-HPN capacity reservation.

    Computing power Quality of Service (QoS)

    ACS supports two computing power QoS types. Use the alibabacloud.com/compute-qos label to set the QoS for a pod.

    QoS type Label Behavior Typical use cases
    Default default Some computing power fluctuations may occur. No forced eviction—failures are resolved through hot migration or user-triggered eviction notifications. Microservices, web applications, compute-intensive tasks
    BestEffort best-effort Some computing power fluctuations may occur. Forced preemption and eviction may occur. You receive an event notification 5 minutes before eviction. Big data computing, audio and video transcoding, batch processing
    ACS QoS types differ from Kubernetes native QoS classes. The Default computing power QoS maps to Kubernetes' Guaranteed QoS class.
    BestEffort instances use dynamic stock. In production, prioritize BestEffort when stock is available and automatically fall back to Default QoS when BestEffort stock runs out. For details, see Custom resource scheduling policies.

    The following example sets the computing power QoS to default:

    apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
        alibabacloud.com/compute-qos: default
    spec:
      containers:
      - name: nginx
        image: registry.cn-hangzhou.aliyuncs.com/acs-sample/nginx:latest

    Compute type and QoS compatibility

    Compute type Supported QoS
    General-purpose (general-purpose) Default, BestEffort
    Compute-optimized (performance) Default, BestEffort
    GPU (gpu) Default, BestEffort
    GPU-HPN (gpu-hpn) Default only

    CPU brand

    The general-purpose and compute-optimized compute types support both Intel and AMD CPUs. Specify the CPU vendor using the alibabacloud.com/cpu-vendors annotation on a pod or in the pod template of a workload.

    To use AMD CPUs, submit a ticketsubmit a ticketsubmit a ticketsubmit a ticket to enable whitelist support. Specifying a CPU vendor for other compute types returns an error.
    Value Description
    intel (default) Intel CPU. If the annotation is not set, Intel is used.
    amd AMD CPU. Requires whitelist enablement.
    intel,amd Either Intel or AMD, selected based on inventory. Custom ordering is not supported when multiple values are specified.

    After a pod is created, check the actual CPU brand in the alibabacloud.com/cpu-vendor label of the pod YAML.

    The following example specifies AMD as the CPU vendor:

    apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
        alibabacloud.com/compute-class: general-purpose
        alibabacloud.com/compute-qos: default
      annotations:
        alibabacloud.com/cpu-vendors: amd
    spec:
      containers:
      - name: nginx
        image: registry.cn-hangzhou.aliyuncs.com/acs-sample/nginx:latest
    Warning

    Do not use ACS system tags—such as alibabacloud.com/compute-class, alibabacloud.com/compute-qos, and alibabacloud.com/cpu-vendor—as filter labels in workload matchLabels. The system may modify these labels, causing controllers to frequently recreate pods and affecting application stability.

    Core features

    Feature Description
    Security isolation Each ACS pod runs in a strongly isolated environment backed by lightweight sandboxed container technology. Instances do not affect each other and are distributed across different physical machines during scheduling for high availability.
    CPU, memory, GPU, and ephemeral storage Configure resources.requests and resources.limits per container using standard Kubernetes methods. ACS pod resources equal the sum of all container resources, and ACS automatically normalizes pod specifications. If limits are not specified, the default limit is the sum of all container resources in the normalized pod.
    Image Each time an ACS pod restarts, it pulls container images from a remote registry over the VPC. Public images require a NAT gateway for the VPC. Store images in Container Registry (ACR) to reduce pull time. For ACR private images, ACS supports a passwordless image pull feature.
    Storage Supports cloud disk, Apsara File Storage NAS, Object Storage Service (OSS), and Cloud Parallel File Storage (CPFS) as persistent storage types.
    Network Each ACS pod gets an independent pod IP, occupying one Elastic Network Interface (ENI) on the vSwitch. Pods can be accessed directly by pod IP, via a LoadBalancer Service, via a ClusterIP Service, or through an Elastic IP Address (EIP).
    Log collection Configure environment variables on the pod to collect stdout or file logs and send them to Simple Log Service (SLS).

    Storage details

    Cloud disk

    Supports ESSD (Enhanced SSD), ESSD AutoPL, standard SSD, and ultra disk types. Supports dynamic provisioning as persistent volumes (PVs). For details, see Disk volumes overview and Use dynamically provisioned volumes for cloud disks.

    NAS

    Supports static provisioning with Capacity and Extreme NAS file systems. Supports dynamic provisioning with Capacity NAS (default). For details, see NAS volumes overview.

    OSS

    Supports static provisioning as PVs. For details, see Use OSS static persistent volumes.

    CPFS

    Supports static provisioning as PVs. For details, see Use CPFS static persistent volumes.

    Network details

    In an ACS cluster, connect pods using any of the following methods:

    Resource specifications

    How ACS normalizes pod specifications

    ACS automatically normalizes pod specifications on submission. The normalization logic works as follows:

    1. ACS calculates the maximum cumulative value of .resources.requests or .resources.limits across all containers.

    2. It rounds up to the nearest supported specification.

    3. The normalized specification is recorded in the alibabacloud.com/pod-use-spec annotation (for example, 2-4Gi).

    4. If upward normalization occurs, ACS adjusts the first container's .resources.requests or .resources.limits to ensure full utilization of paid resources.

    If you do not set resources.requests or resources.limits, the default pod specification is 2 vCPUs and 4 GiB of memory.

    Example: If the cumulative value is 2 vCPUs and 3.5 GiB of memory, ACS normalizes the pod to 2 vCPUs and 4 GiB. The extra 0.5 GiB is allocated to the first container, and the pod gets the annotation alibabacloud.com/pod-use-spec=2-4Gi.

    Before normalization:

    apiVersion: v1
kind: Pod
metadata:
  labels:
    app: nginx
    alibabacloud.com/compute-class: general-purpose
    alibabacloud.com/compute-qos: default
  name: nginx
spec:
  containers:
  - name: nginx
    image: anolis-registry.cn-zhangjiakou.cr.aliyuncs.com/openanolis/nginx:1.14.1-8.6
    ports:
    - containerPort: 80
    resources:
      requests:
        cpu: 2         # 2 vCPUs
        memory: "3.5Gi"
        ephemeral-storage: "30Gi"

    After normalization:

    apiVersion: v1
kind: Pod
metadata:
  annotations:
    alibabacloud.com/pod-use-spec: "2-4Gi"
  labels:
    app: nginx
    alibabacloud.com/compute-class: general-purpose
    alibabacloud.com/compute-qos: default
  name: nginx
spec:
  containers:
  - name: nginx
    image: anolis-registry.cn-zhangjiakou.cr.aliyuncs.com/openanolis/nginx:1.14.1-8.6
    ports:
    - containerPort: 80
    resources:
      requests:
        cpu: 2         # 2 vCPUs
        memory: "4Gi"  # Normalized from 3.5 GiB to 4 GiB
        ephemeral-storage: "30Gi"

    Specify pod specifications using an annotation

    For Burstable QoS workloads (.resources.limits > .resources.requests), use the alibabacloud.com/pod-required-spec: "X-YGi" annotation to declare the target pod specification explicitly.

    Scope

    Format

    Use <CPU>-<Memory> format, where CPU is in cores (for example, "2" means 2 vCPUs) and memory is in GiB (for example, "4Gi" means 4 GiB).

    Validation rules

    1. Invalid format (missing units, using Mi, reversed order) causes pod creation to fail.

    2. If the annotation is set but no container .resources are defined, the system normalizes strictly to the annotation value and does not fall back to defaults (2 vCPUs / 4 GiB).

    3. If the annotation value is less than the sum of all container .resources.requests, pod creation fails.

    4. If the annotation value exceeds the sum of all container .resources.limits, the annotation value is used as the target normalized specification. In multi-container pods, the first container (the primary container) receives the difference between the annotation value and the current sum of limits, with .resources.requests adjusted accordingly.

    Example

    Setting alibabacloud.com/pod-required-spec: "2-4Gi" when container resources declare 1 vCPU / 2 GiB requests and 2 vCPU / 3.5 GiB limits causes ACS to normalize the pod to 2 vCPUs and 4 GiB.

    Before normalization (.resources.limits.memory is 3.5 GiB):

    apiVersion: v1
kind: Pod
metadata:
  labels:
    app: nginx
    alibabacloud.com/compute-class: general-purpose
    alibabacloud.com/compute-qos: default
  annotations:
    alibabacloud.com/pod-required-spec: "2-4Gi"
  name: nginx
spec:
  containers:
  - name: nginx
    image: anolis-registry.cn-zhangjiakou.cr.aliyuncs.com/openanolis/nginx:1.14.1-8.6
    ports:
    - containerPort: 80
    resources:
      requests:
        cpu: 1         # 1 vCPU
        memory: "2Gi"
        ephemeral-storage: "30Gi"
      limits:
        cpu: 2         # 2 vCPUs
        memory: "3.5Gi"
        ephemeral-storage: "30Gi"

    After normalization (.resources.limits.memory is normalized from 3.5 GiB to 4 GiB):

    apiVersion: v1
kind: Pod
metadata:
  annotations:
    alibabacloud.com/pod-required-spec: "2-4Gi"
    alibabacloud.com/pod-use-spec: "2-4Gi"
  labels:
    app: nginx
    alibabacloud.com/compute-class: general-purpose
    alibabacloud.com/compute-qos: default
  name: nginx
spec:
  containers:
  - name: nginx
    image: anolis-registry.cn-zhangjiakou.cr.aliyuncs.com/openanolis/nginx:1.14.1-8.6
    ports:
    - containerPort: 80
    resources:
      requests:
        cpu: 1         # 1 vCPU
        memory: "2Gi"
        ephemeral-storage: "30Gi"
      limits:
        cpu: 2         # 2 vCPUs
        memory: "4Gi"  # Normalized from 3.5 GiB to 4 GiB
        ephemeral-storage: "30Gi"

    General-purpose compute type

    Important

    To use a pod with more than 16 vCPUs or more than 128 GiB of memory, submit a ticketsubmit a ticketsubmit a ticketsubmit a ticket to request approval.

    vCPU Memory (GiB) Memory step size (GiB) Network bandwidth (egress + ingress) (Gbps) Storage
    0.25 0.5, 1, 2 N/A 0.08 Up to 30 GiB is free. Storage exceeding 30 GiB is billed for the excess. Maximum: 512 GiB. Mount persistent volumes (such as NAS) for additional storage.
    0.5 1–4 1 0.08
    1 1–8 0.1
    1.5 2–12 1
    2 2–16
    2.5 3–20 1.5
    3 3–24
    3.5 4–28
    4 4–32
    4.5 5–36
    5 5–40
    5.5 6–44
    6 6–48
    6.5 7–52 2.5
    7 7–56
    7.5 8–60
    8 8–64
    8.5 9–68
    9 9–72
    9.5 10–76
    10 10–80
    10.5 11–84
    11 11–88
    11.5 12–92
    12 12–96
    12.5 13–100 3
    13 13–104
    13.5 14–108
    14 14–112
    14.5 15–116
    15 15–120
    15.5 16–124
    16 16–128
    24 24, 48, 96, 192 N/A 4.5
    32 32, 64, 128, 256 N/A 6
    48 48, 96, 192, 384 N/A 12.5
    64 64, 128, 256, 512 N/A 20

    Compute-optimized compute type

    Important

    To use a pod with more than 16 vCPUs or more than 128 GiB of memory, submit a ticketsubmit a ticketsubmit a ticketsubmit a ticket to request approval.

    vCPU Memory (GiB) Memory step size (GiB) Network bandwidth (egress + ingress) (Gbps) Storage
    0.25 0.5, 1, 2 N/A 0.1 Up to 30 GiB is free. Storage exceeding 30 GiB is billed for the excess. Maximum: 512 GiB. Mount persistent volumes (such as NAS) for additional storage.
    0.5 1–4 1 0.5
    1 1–8
    1.5 2–12
    2 2–16 1.5
    2.5 3–20
    3 3–24
    3.5 4–28
    4 4–32 2
    4.5 5–36
    5 5–40
    5.5 6–44
    6 6–48 2.5
    6.5 7–52
    7 7–56
    7.5 8–60
    8 8–64 3
    8.5 9–68
    9 9–72
    9.5 10–76
    10 10–80 3.5
    10.5 11–84
    11 11–88
    11.5 12–92
    12 12–96 4
    12.5 13–100
    13 13–104
    13.5 14–108
    14 14–112 4.5
    14.5 15–116
    15 15–120
    15.5 16–124
    16 16–128 6
    24 24, 48, 96, 192 N/A 8
    32 32, 64, 128, 256 N/A 10
    48 48, 96, 192, 384 N/A 16
    64 64, 128, 256, 512 N/A 25

    Accelerated compute types

    The following GPU card types are supported. Specifications vary by card type. To get the exact specification mapping, submit a ticketsubmit a ticketsubmit a ticketsubmit a ticket.

    Important

    • For specifications with 16 GiB or less of memory, ACS covers memory overhead. For specifications exceeding 16 GiB, memory overhead is distributed to the corresponding pods. Reserve sufficient resources for your applications.

    • System disk capacity of 30 GiB or less (including image size) does not incur additional charges. You are billed only for capacity exceeding 30 GiB.

    Warning

    In an ACS cluster, GPU and GPU-HPN pod specifications are automatically normalized to Guaranteed QoS (request equals limit) on submission. When accessing ACS GPU compute through other channels—such as ACK or ACK One clusters—normalization is not reflected in pod metadata. Make sure the pod QoS remains unchanged before and after submission (GPU compute types must maintain Guaranteed QoS) to prevent pod status update failures.

    GU8TF

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (96 GiB GPU memory) 2 2–16 1 30–256
    4 4–32 1
    6 6–48 1
    8 8–64 1
    10 10–80 1
    12 12–96 1
    14 14–112 1
    16 16–128 1
    22 22, 32, 64, 128 N/A
    2 (96 GiB × 2 GPU memory) 16 16–128 1 30–512
    32 32, 64, 128, 230 N/A
    46 64, 128, 230 N/A
    4 (96 GiB × 4 GPU memory) 32 32, 64, 128, 256 N/A 30–1024
    64 64, 128, 256, 460 N/A
    92 128, 256, 460 N/A
    8 (96 GiB × 8 GPU memory) 64 64, 128, 256, 512 N/A 30–2048
    128 128, 256, 512, 920 N/A
    184 256, 512, 920 N/A

    GU8TEF

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (141 GiB GPU memory) 2 2–16 1 30–768
    4 4–32 1
    6 6–48 1
    8 8–64 1
    10 10–80 1
    12 12–96 1
    14 14–112 1
    16 16–128 1
    22 22, 32, 64, 128, 225 N/A
    2 (141 GiB × 2 GPU memory) 16 16–128 1 30–1536
    32 32, 64, 128, 256 N/A
    46 64, 128, 256, 450 N/A
    4 (141 GiB × 4 GPU memory) 32 32, 64, 128, 256 N/A 30–3072
    64 64, 128, 256, 512 N/A
    92 128, 256, 512, 900 N/A
    8 (141 GiB × 8 GPU memory) 64 64, 128, 256, 512 N/A 30–6144
    128 128, 256, 512, 1024 N/A
    184 256, 512, 1024, 1800 N/A

    L20 (GN8IS)

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (48 GiB GPU memory) 2 2–16 1 30–256
    4 4–32 1
    6 6–48 1
    8 8–64 1
    10 10–80 1
    12 12–96 1
    14 14–112 1
    16 16–120 1
    2 (48 GiB × 2 GPU memory) 16 16–128 1 30–512
    32 32, 64, 128, 230 N/A
    4 (48 GiB × 4 GPU memory) 32 32, 64, 128, 256 N/A 30–1024
    64 64, 128, 256, 460 N/A
    8 (48 GiB × 8 GPU memory) 64 64, 128, 256, 512 N/A 30–2048
    128 128, 256, 512, 920 N/A

    L20X (GX8SF)

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    8 (141 GiB × 8 GPU memory) 184 1800 N/A 30–6144

    P16EN

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (96 GiB GPU memory) 2 2–16 1 30–384
    4 4–32 1
    6 6–48 1
    8 8–64 1
    10 10–80 1
    2 (96 GiB × 2 GPU memory) 4 4–32 1 30–768
    6 6–48 1
    8 8–64 1
    16 16–128 1
    22 32, 64, 128, 225 N/A
    4 (96 GiB × 4 GPU memory) 8 8–64 1 30–1536
    16 16–128 1
    32 32, 64, 128, 256 N/A
    46 64, 128, 256, 450 N/A
    8 (96 GiB × 8 GPU memory) 16 16–128 1 30–3072
    32 32, 64, 128, 256 N/A
    64 64, 128, 256, 512 N/A
    92 128, 256, 512, 900 N/A
    16 (96 GiB × 16 GPU memory) 32 32, 64, 128, 256 N/A 30–6144
    64 64, 128, 256, 512 N/A
    128 128, 256, 512, 1024 N/A
    184 256, 512, 1024, 1800 N/A

    G49E

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (48 GiB GPU memory) 2 2–16 1 30–256
    4 4–32 1
    6 6–48 1
    8 8–64 1
    10 10–80 1
    12 12–96 1
    14 14–112 1
    16 16–120 1
    2 (48 GiB × 2 GPU memory) 16 16–128 1 30–512
    32 32, 64, 128, 230 N/A
    4 (48 GiB × 4 GPU memory) 32 32, 64, 128, 256 N/A 30–1024
    64 64, 128, 256, 460 N/A
    8 (48 GiB × 8 GPU memory) 64 64, 128, 256, 512 N/A 30–2048
    128 128, 256, 512, 920 N/A

    T4

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (16 GiB GPU memory) 2 2–8 1 30–1536
    4 4–16 1
    6 6–24 1
    8 8–32 1
    10 10–40 1
    12 12–48 1
    14 14–56 1
    16 16–64 1
    24 24, 48, 90 N/A
    2 (16 GiB × 2 GPU memory) 16 16–64 1 30–1536
    24 24, 48, 96 N/A
    32 32, 64, 128 N/A
    48 48, 96, 180 N/A

    A10

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB)
    1 (24 GiB GPU memory) 2 2–8 1 30–256
    4 4–16 1
    6 6–24 1
    8 8–32 1
    10 10–40 1
    12 12–48 1
    14 14–56 1
    16 16–60 1
    2 (24 GiB × 2 GPU memory) 16 16–64 1 30–512
    32 32, 64, 120 N/A
    4 (24 GiB × 4 GPU memory) 32 32, 64, 128 N/A 30–1024
    64 64, 128, 240 N/A
    8 (24 GiB × 8 GPU memory) 64 64, 128, 256 N/A 30–2048
    128 128, 256, 480 N/A

    G59

    GPU (cards) vCPU Memory (GiB) Memory step size (GiB) Storage (GiB) Networking
    1 (32 GiB GPU memory) 2 2–16 1 30–256 1 Gbps per vCPU
    4 4–32 1
    6 6–48 1
    8 8–64 1
    10 10–80 1
    12 12–96 1
    14 14–112 1
    16 16–128 1
    22 22, 32, 64, 128 N/A
    2 (32 GiB × 2 GPU memory) 16 16–128 1 30–512
    32 32, 64, 128, 256 N/A
    46 64, 128, 256, 360 N/A
    4 (32 GiB × 4 GPU memory) 32 32, 64, 128, 256 N/A 30–1024
    64 64, 128, 256, 512 N/A
    92 128, 256, 512, 720 N/A
    8 (32 GiB × 8 GPU memory) 64 64, 128, 256, 512 N/A 30–2048
    128 128, 256, 512, 1024 N/A 100 Gbps
    184 256, 512, 1024, 1440 N/A

    Automatic specification normalization for GPU pods

    If you do not set resource requests or limits for a GPU pod, the pod selects the smallest supported specification for the GPU type—for example, 2 vCPUs, 2 GiB memory, and 1 GPU card.

    For unsupported specifications, ACS normalizes automatically. After normalization:

    • The container's .resources.requests remain unchanged.

    • The normalized specification appears in the alibabacloud.com/pod-use-spec annotation.

    • If a container's .resources.limits exceeds the normalized specification, ACS sets the container's limit to match the pod specification.

    CPU and memory normalization: If the sum across all containers is 2 vCPUs and 3.5 GiB, ACS normalizes to 2 vCPUs and 4 GiB. The extra resources apply to the first container, and the pod gets the annotation alibabacloud.com/pod-use-spec=2-4Gi. If a single container specifies 3 vCPUs and 5 GiB as limits, that container's limit is set to 2 vCPUs and 5 GiB.
    GPU normalization: If the GPU count requested is not supported, pod submission fails.

    GPU-HPN specifications

    For GPU-HPN instances, ACS enforces resource alignment by setting limit equal to request. Pod resource specifications are also constrained by node capacity—if a pod's specifications exceed node capacity, the pod enters a pending state due to insufficient resources. For details about node specifications, see the purchasing specifications.

    Limitations

    ACS uses virtual nodes and does not run pods on standard nodes. Because pods are distributed across Alibaba Cloud's global resource pool, certain Kubernetes features are not supported.

    Limitation Description System response Alternative
    DaemonSet DaemonSet workloads are restricted. Pod starts but does not function normally. Use the sidecar pattern to deploy multiple containers in a pod.
    NodePort Service Host port mapping to containers is not supported. Submission rejected. Use a LoadBalancer type Service (type=LoadBalancer).
    HostNetwork Host port mapping to containers is restricted. Automatically rewritten to HostNetwork=false. Not required.
    HostIPC Inter-process communication between container and host processes is restricted. Automatically rewritten to HostIPC=false. Not required.
    HostPID Container access to the host's PID namespace is restricted. Automatically rewritten to HostPID=false. Not required.
    HostUsers User namespace usage is restricted. Automatically rewritten to an empty value. Not required.
    DNSPolicy Only None, Default, and ClusterFirst are allowed. ClusterFirstWithHostNet is rewritten to ClusterFirst. All other policies are rejected. Use only allowed values.
    Container environment variable format For GPU and GPU-HPN compute types, environment variable names must contain only letters, numbers, underscores, dots, or hyphens. The first character cannot be a number. Pod startup fails. Use environment variable names that meet these requirements.
    Number of container environment variables Limited to approximately 2,000 per container due to Linux system constraints. If enableServiceLinks is enabled (default: true), service information from all services in the namespace is injected as environment variables, which may exceed the limit. Pod startup fails. Reduce environment variable count. For deployments with many services, set enableServiceLinks: false on the pod.

    Reserved ports

    Avoid using the following ports when deploying services in ACS.

    Port Used by
    111, 10250, 10255 ACS cluster interfaces, including exec, logs, and metrics.