Fix deployment on arm64 architecture (#109)

abuccts · web-flow · commit 7e1a26301686 · 2025-11-07T16:35:34.000-08:00
Fix service build and deployment on arm64 architecture.
diff --git a/.github/workflows/build-deploy-changes.yaml b/.github/workflows/build-deploy-changes.yaml
@@ -23,6 +23,8 @@ jobs:
       image: ubuntu:latest
       volumes:
         - /var/run/docker.sock:/var/run/docker.sock
+    env:
+      DOCKER_BUILDKIT: "1"
     steps:
       - name: Install git
         run: |
@@ -76,8 +78,9 @@ jobs:
       - name: Install Package
         if: steps.check.outputs.has_changed == 'true'
         run: |
-          DEBIAN_FRONTEND=noninteractive apt install -y python3 python-is-python3 pip git unzip docker-cli ca-certificates curl apt-transport-https lsb-release gnupg parallel
+          DEBIAN_FRONTEND=noninteractive apt install -y python3 python-is-python3 pip git unzip ca-certificates curl apt-transport-https lsb-release gnupg parallel
           curl -sL https://aka.ms/InstallAzureCLIDeb | bash
+          curl -fsSL https://get.docker.com | sh
 
       - name: Install python libs
         if: steps.check.outputs.has_changed == 'true'
@@ -98,6 +101,11 @@ jobs:
           mv $GITHUB_WORKSPACE/config/auth-configuration /tmp/
           ls -l /tmp/auth-configuration
 
+      - name: Login to GHCR
+        if: steps.check.outputs.has_changed == 'true'
+        run: |
+          docker login ghcr.io -u ${{ github.actor }} -p ${{ secrets.GITHUB_TOKEN }}
+
       - name: Build Images of Changed Services
         if: steps.check.outputs.has_changed == 'true'
         run: |
diff --git a/contrib/kubespray/script/environment.sh b/contrib/kubespray/script/environment.sh
@@ -51,5 +51,5 @@ sudo python3 -m pip install -r ${HOME}/pai-deploy/kubespray/requirements.txt
 
 # workaround python3-apt issue
 SOABI=$(python3 -c 'import sysconfig; print(sysconfig.get_config_var("SOABI"))')
-sudo ln -s /usr/lib/python3/dist-packages/apt_inst.${SOABI}.so /usr/lib/python3/dist-packages/apt_inst.so
-sudo ln -s /usr/lib/python3/dist-packages/apt_pkg.${SOABI}.so /usr/lib/python3/dist-packages/apt_pkg.so
+sudo ln -sf /usr/lib/python3/dist-packages/apt_inst.${SOABI}.so /usr/lib/python3/dist-packages/apt_inst.so
+sudo ln -sf /usr/lib/python3/dist-packages/apt_pkg.${SOABI}.so /usr/lib/python3/dist-packages/apt_pkg.so
diff --git a/src/device-plugin/build/k8s-host-device-plugin.k8s.dockerfile b/src/device-plugin/build/k8s-host-device-plugin.k8s.dockerfile
@@ -0,0 +1,6 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+FROM everpeace/k8s-host-device-plugin
+
+#TODO: add arm64 image
diff --git a/src/device-plugin/build/k8s-nvidia-device-plugin.k8s.dockerfile b/src/device-plugin/build/k8s-nvidia-device-plugin.k8s.dockerfile
@@ -0,0 +1,4 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+FROM nvcr.io/nvidia/k8s-device-plugin:v0.18.0
diff --git a/src/device-plugin/build/k8s-rdma-shared-dev-plugin.k8s.dockerfile b/src/device-plugin/build/k8s-rdma-shared-dev-plugin.k8s.dockerfile
@@ -0,0 +1,4 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+FROM ghcr.io/mellanox/k8s-rdma-shared-dev-plugin:1.4.0
diff --git a/src/device-plugin/build/k8s-rocm-device-plugin.k8s.dockerfile b/src/device-plugin/build/k8s-rocm-device-plugin.k8s.dockerfile
@@ -0,0 +1,6 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+FROM rocm/k8s-device-plugin
+
+# no arm64 support
diff --git a/src/device-plugin/deploy/device-plugin.yaml.template b/src/device-plugin/deploy/device-plugin.yaml.template
@@ -29,16 +29,15 @@ spec:
       name: k8s-host-device-plugin-ds
   template:
     metadata:
-      annotations:
-        scheduler.alpha.kubernetes.io/critical-pod: ""
       labels:
         name: k8s-host-device-plugin-ds
     spec:
+      priorityClassName: pai-daemon-priority
       tolerations:
       - key: CriticalAddonsOnly
         operator: Exists
       containers:
-      - image: luciaopenai.azurecr.io/luciaopenai/k8s-host-device-plugin:latest
+      - image: {{ cluster_cfg["cluster"]["docker-registry"]["prefix"] }}k8s-host-device-plugin:{{ cluster_cfg["cluster"]["docker-registry"]["tag"] }}
         name: k8s-host-device-plugin-ctr
         securityContext:
           privileged: true
@@ -62,3 +61,5 @@ spec:
             items:
             - key: config.json
               path: config.json
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
diff --git a/src/device-plugin/deploy/service.yaml b/src/device-plugin/deploy/service.yaml
@@ -25,6 +25,7 @@ prerequisite:
 template-list:
   - start.sh
   - delete.sh
+  - device-plugin.yaml
 
 start-script: start.sh
 stop-script: stop.sh
diff --git a/src/device-plugin/deploy/start.sh.template b/src/device-plugin/deploy/start.sh.template
@@ -29,25 +29,33 @@ pushd $(dirname "$0") > /dev/null
 # Begin: NVIDIA GPU device plugin
 {% if 'nvidia.com/gpu' in cluster_cfg['device-plugin']['devices'] %}
 
-curl -s https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.15.0/deployments/static/nvidia-device-plugin.yml \
-| sed 's|nvcr.io/nvidia/k8s-device-plugin|luciaopenai.azurecr.io/luciaopenai/nvidia/k8s-device-plugin|' \
+{ curl -s https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.15.0/deployments/static/nvidia-device-plugin.yml \
+| sed 's|nvcr.io/nvidia/k8s-device-plugin:v0.15.0|{{ cluster_cfg['cluster']['docker-registry']['prefix'] }}k8s-nvidia-device-plugin:{{ cluster_cfg['cluster']['docker-registry']['tag'] }}|' \
 | sed -E '/^[[:space:]]*allowPrivilegeEscalation: false/ {
     h
     s/^([[:space:]]*)allowPrivilegeEscalation: false.*$/\1privileged: false/
     G
     s/(^[[:space:]]*allowPrivilegeEscalation: false.*)\n([[:space:]]*privileged: false)/\1\n\2/
-}' \
-    | kubectl apply --overwrite=true -f - || exit $?
+}';
+  cat <<'YAML'
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
+YAML
+} | kubectl apply --overwrite=true -f - || exit $?
 
 {% endif %}
 # End: NVIDIA GPU device plugin
 
 # Begin: AMD GPU device plugin
 {% if 'amd.com/gpu' in cluster_cfg['device-plugin']['devices'] %}
 
-curl -s https://raw.githubusercontent.com/ROCm/k8s-device-plugin/master/k8s-ds-amdgpu-dp.yaml \
-| sed 's|rocm/k8s-device-plugin|luciaopenai.azurecr.io/luciaopenai/rocm/k8s-device-plugin:{{ cluster_cfg['cluster']['docker-registry']['tag'] }}|' \
-    | kubectl apply --overwrite=true -f - || exit $?
+{ curl -s https://raw.githubusercontent.com/ROCm/k8s-device-plugin/master/k8s-ds-amdgpu-dp.yaml \
+| sed 's|rocm/k8s-device-plugin|{{ cluster_cfg['cluster']['docker-registry']['prefix'] }}k8s-rocm-device-plugin:{{ cluster_cfg['cluster']['docker-registry']['tag'] }}|';
+  cat <<'YAML'
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
+YAML
+} | kubectl apply --overwrite=true -f - || exit $?
 
 {% endif %}
 # End: AMD GPU device plugin
@@ -75,9 +83,13 @@ kubectl apply --overwrite=true -f device-plugin.yaml || exit $?
 {% if 'rdma/hca' in cluster_cfg['device-plugin']['devices'] %}
 
 kubectl apply --overwrite=true -f rdma-devices.yaml || exit $?
-curl -s https://raw.githubusercontent.com/Mellanox/k8s-rdma-shared-dev-plugin/v1.4.0/deployment/k8s/base/daemonset.yaml \
-    | sed 's|ghcr.io/mellanox/k8s-rdma-shared-dev-plugin|luciaopenai.azurecr.io/luciaopenai/k8s-rdma-shared-dev-plugin:{{ cluster_cfg['cluster']['docker-registry']['tag'] }}|' \
-    | kubectl apply --overwrite=true -f - || exit $?
+{ curl -s https://raw.githubusercontent.com/Mellanox/k8s-rdma-shared-dev-plugin/v1.4.0/deployment/k8s/base/daemonset.yaml \
+| sed 's|ghcr.io/mellanox/k8s-rdma-shared-dev-plugin|{{ cluster_cfg['cluster']['docker-registry']['prefix'] }}k8s-rdma-shared-dev-plugin:{{ cluster_cfg['cluster']['docker-registry']['tag'] }}|';
+  cat <<'YAML'
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
+YAML
+} | kubectl apply --overwrite=true -f - || exit $?
 
 {% endif %}
 
diff --git a/src/frameworkcontroller/deploy/frameworkcontroller.yaml.template b/src/frameworkcontroller/deploy/frameworkcontroller.yaml.template
@@ -51,3 +51,5 @@ spec:
       - name: frameworkcontroller-config
         configMap:
           name: frameworkcontroller-config
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
diff --git a/src/grafana/build/grafana.common.dockerfile b/src/grafana/build/grafana.common.dockerfile
@@ -18,6 +18,7 @@
 
 FROM ubuntu:22.04
 
+ARG TARGETARCH
 ENV \
   GRAFANA_VERSION=10.4.18+security~01 \
   GF_PLUGIN_DIR=/grafana-plugins \
@@ -29,7 +30,7 @@ ENV \
 RUN \
   apt-get update && \
   apt-get -y --force-yes --no-install-recommends install libfontconfig wget ca-certificates adduser libfontconfig1 musl curl jq && \
-  wget -O /tmp/grafana.deb https://dl.grafana.com/oss/release/grafana_${GRAFANA_VERSION}_amd64.deb && \
+  wget -O /tmp/grafana.deb https://dl.grafana.com/oss/release/grafana_${GRAFANA_VERSION}_${TARGETARCH}.deb && \
   dpkg -i /tmp/grafana.deb && \
   rm -f /tmp/grafana.deb && \
   ### branding && \
diff --git a/src/hivedscheduler/build/hivedscheduler.common.dockerfile b/src/hivedscheduler/build/hivedscheduler.common.dockerfile
diff --git a/src/hivedscheduler/build/hivedscheduler.k8s.dockerfile b/src/hivedscheduler/build/hivedscheduler.k8s.dockerfile
@@ -0,0 +1,30 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+FROM golang:1.24.3-alpine3.21 AS builder
+
+ARG TEST=false
+ENV GOPATH=/go
+ENV PROJECT_DIR=/src
+ENV INSTALL_DIR=/opt/hivedscheduler/hivedscheduler
+
+RUN apk update && apk add --no-cache bash
+RUN mkdir -p ${PROJECT_DIR} ${INSTALL_DIR}
+COPY src ${PROJECT_DIR}
+RUN if [ ${TEST} == "true" ]; \
+  then ${PROJECT_DIR}/build/hivedscheduler/go-build.sh test; \
+  else ${PROJECT_DIR}/build/hivedscheduler/go-build.sh; fi && \
+  mv ${PROJECT_DIR}/dist/hivedscheduler/* ${INSTALL_DIR}
+
+
+FROM alpine:3.21
+
+ENV INSTALL_DIR=/opt/hivedscheduler/hivedscheduler
+
+RUN apk update && apk add --no-cache bash
+RUN apk upgrade --no-cache
+
+COPY --from=builder ${INSTALL_DIR} ${INSTALL_DIR}
+WORKDIR ${INSTALL_DIR}
+
+ENTRYPOINT ["./start.sh"]
diff --git a/src/hivedscheduler/build/kube-scheduler.k8s.dockerfile b/src/hivedscheduler/build/kube-scheduler.k8s.dockerfile
@@ -0,0 +1,4 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+FROM registry.k8s.io/kube-scheduler:v1.28.9
diff --git a/src/hivedscheduler/deploy/hivedscheduler.yaml.template b/src/hivedscheduler/deploy/hivedscheduler.yaml.template
@@ -53,6 +53,8 @@ spec:
       - name: hivedscheduler-config
         configMap:
           name: hivedscheduler-config
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
 
 {%- for vc in cluster_cfg['hivedscheduler']['structured-config']['virtualClusters'] %}
 ---
@@ -96,4 +98,6 @@ spec:
       - name: hivedscheduler-config
         configMap:
           name: hivedscheduler-config
+      imagePullSecrets:
+      - name: {{ cluster_cfg["cluster"]["docker-registry"]["secret-name"] }}
 {%- endfor %}
diff --git a/src/job-exporter/build/job-exporter.common.dockerfile b/src/job-exporter/build/job-exporter.common.dockerfile
@@ -17,6 +17,8 @@
 
 
 FROM mcr.microsoft.com/mirror/nvcr/nvidia/cuda:12.0.1-runtime-ubuntu22.04
+
+ARG TARGETARCH
 # Register the ROCM package repository, and install rocm-dev package
 ARG ROCM_VERSION=6.2.2
 ARG AMDGPU_VERSION=6.2.2
@@ -25,8 +27,8 @@ RUN echo "$APT_PREF" > /etc/apt/preferences.d/rocm-pin-600
 
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends curl libnuma-dev gnupg \
   && curl -sL https://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - \
-  && printf "deb [arch=amd64] https://repo.radeon.com/rocm/apt/$ROCM_VERSION/ jammy main" | tee /etc/apt/sources.list.d/rocm.list \
-  && printf "deb [arch=amd64] https://repo.radeon.com/amdgpu/$AMDGPU_VERSION/ubuntu jammy main" | tee /etc/apt/sources.list.d/amdgpu.list \
+  && printf "deb https://repo.radeon.com/rocm/apt/$ROCM_VERSION/ jammy main" | tee /etc/apt/sources.list.d/rocm.list \
+  && printf "deb https://repo.radeon.com/amdgpu/$AMDGPU_VERSION/ubuntu jammy main" | tee /etc/apt/sources.list.d/amdgpu.list \
   && apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
   sudo \
   libelf1 \
@@ -74,13 +76,12 @@ COPY build/update-dcgm.py .
 # For the job exporter
 ENV NERDCTL_VERSION=2.1.3
 RUN apt-get update && apt-get install --no-install-recommends -y wget ca-certificates
-RUN wget https://github.com/containerd/nerdctl/releases/download/v${NERDCTL_VERSION}/nerdctl-${NERDCTL_VERSION}-linux-amd64.tar.gz && \
+RUN wget -O /tmp/nerdctl.tar.gz https://github.com/containerd/nerdctl/releases/download/v${NERDCTL_VERSION}/nerdctl-${NERDCTL_VERSION}-linux-${TARGETARCH}.tar.gz && \
     mkdir -p /tmp/nerdctl && \
-    tar -xzvf nerdctl-${NERDCTL_VERSION}-linux-amd64.tar.gz -C /tmp/nerdctl && \
+    tar -xzvf /tmp/nerdctl.tar.gz -C /tmp/nerdctl && \
     mv /tmp/nerdctl/nerdctl /usr/local/bin/nerdctl && \
     mkdir -p /job_exporter && \
-    rm -rf /tmp/nerdctl && \
-    rm -rf nerdctl-${NERDCTL_VERSION}-linux-amd64.tar.gz
+    rm -rf /tmp/nerdctl*
 
 COPY requirements.txt /job_exporter/
 RUN pip3 install -r /job_exporter/requirements.txt
diff --git a/src/openpai-runtime/build/openpai-runtime.common.dockerfile b/src/openpai-runtime/build/openpai-runtime.common.dockerfile
@@ -64,6 +64,7 @@ WORKDIR /kube-runtime/src
 COPY src/src ./
 COPY src/requirements.txt ./
 
+#TODO: update the hardcode image for arm64
 COPY --from=frameworkcontroller/frameworkbarrier:v1.0.0 $BARRIER_DIR/frameworkbarrier ./init.d
 COPY --from=builder ${INSTALL_DIR}/* ./runtime.d/
 
