starlingx/integ
Code Issues Proposed changes

integ: add nvidia gpu-operator helm charts

Browse Source 
This commit adds nvidia gpu-operator helm charts use case for
custom container runtime feature. To load nvidia-gpu-operator
on starlingx:

system service-parameter-add platform container_runtime \
custom_container_runtime=\
nvidia:/usr/local/nvidia/toolkit/nvidia-container-runtime

And define  runtimeClass for nvidia gpu  pods:

kind: RuntimeClass
apiVersion: node.k8s.io/v1beta1
metadata:
  name: nvidia
handler: nvidia

The above will direct all containerd creations of pods with nvidia
runtimeClass to nvidia-container-runtime -- where the nvidia-conta
iner-runtime is installed by the operator onto a hostMount.

Story: 2008434
Task: 41978

Signed-off-by: Babak Sarashki <babak.sarashki@windriver.com>
Change-Id: Ifea8cdf6eb89a159f446c53566279e72fcf0e45e
This commit is contained in:
Babak Sarashki 2021-03-03 12:15:52 +00:00 committed by bsarashk
parent 2d490aaf6a
commit 7badc1dad1
6 changed files with 733 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
centos_pkg_dirs
View File

@ -85,3 +85,4 @@ python/python-webencodings
python/python-daemon
base/inih
base/pf-bb-config
gpu/gpu-operator

1
centos_tarball-dl.lst
View File

@ -71,3 +71,4 @@ xxHash-1f40c6511fa8dd9d2e337ca8c9bc18b3e87663c9.tar.gz#xxHash#https://api.github
zstd-f4340f46b2387bc8de7d5320c0b83bb1499933ad.tar.gz#zstd#https://api.github.com/repos/facebook/zstd/tarball/f4340f46b2387bc8de7d5320c0b83bb1499933ad#https##
inih-b1dbff4b0bd1e1f40d237e21011f6dee0ec2fa69.tar.gz#inih-44#https://github.com/benhoyt/inih/tarball/b1dbff4b0bd1e1f40d237e21011f6dee0ec2fa69#https##
pf-bb-config-945712e8876be2003f2f31de70353c48501519fa.tar.gz#pf-bb-config-21.3#https://github.com/intel/pf-bb-config/tarball/945712e8876be2003f2f31de70353c48501519fa#https##
gpu-operator-1.6.0.tar.gz#gpu-operator-1.6.0#https://github.com/NVIDIA/gpu-operator/archive/1.6.0.tar.gz##https##

8
gpu/gpu-operator/centos/build_srpm.data Normal file
View File

@ -0,0 +1,8 @@
VERSION=1.6.0
TAR_NAME=gpu-operator
TAR="$TAR_NAME-$VERSION.tar.gz"
COPY_LIST=" \
$PKG_BASE/files/* \
$STX_BASE/downloads/$TAR"
TIS_PATCH_VER=PKG_GITREVCOUNT

46
gpu/gpu-operator/centos/gpu-operator.spec Normal file
View File

@ -0,0 +1,46 @@
# Build variables
%global helm_ver v3
%global helm_folder /usr/lib/helm
Summary: StarlingX nvidia gpu-operator helm chart
Name: gpu-operator
Version: 1.6.0
Release: 0%{?_tis_dist}.%{tis_patch_ver}
License: Apache-2.0
Group: base
Packager: Wind River <info@windriver.com>
URL: https://github.com/NVIDIA/gpu-operator/tree/gh-pages
Source0: %{name}-%{version}.tar.gz
BuildArch: noarch
Patch01: deployments-setup-configmap-with-assets-for-volumemo.patch
Patch02: enablement-support-on-starlingx-cloud-platform.patch
BuildRequires: helm
%define debug_package %{nil}
%description
StarlingX port of NVIDIA gpu-operator
%prep
%setup
%patch01 -p1
%patch02 -p1
%build
cp -r assets deployments/gpu-operator/assets
helm lint deployments/gpu-operator
mkdir build_results
helm package --version %{helm_ver}-%{version}.%{tis_patch_ver} --app-version %{version} -d build_results deployments/gpu-operator
%install
install -d -m 755 ${RPM_BUILD_ROOT}%{helm_folder}
install -p -D -m 755 build_results/%{name}-%{helm_ver}-%{version}.%{tis_patch_ver}.tgz ${RPM_BUILD_ROOT}%{helm_folder}
%files
%defattr(-,root,root,-)
%{helm_folder}

137
gpu/gpu-operator/files/deployments-setup-configmap-with-assets-for-volumemo.patch Normal file
View File

@ -0,0 +1,137 @@
From b968c69971a195aba4e0c03e8a70df074c128f69 Mon Sep 17 00:00:00 2001
From: Babak Sarashki <babak.sarashki@windriver.com>
Date: Sat, 6 Mar 2021 00:22:40 +0000
Subject: [PATCH 1/2] deployments: setup configmap with assets for volumemounts
This feature allows inclusion of assets/ in the helm chart and their
export to the gpu-operator pod through configmap volumeMounts.
Signed-off-by: Babak Sarashki <babak.sarashki@windriver.com>
---
.../gpu-operator/templates/operator.yaml | 45 +++++++++++++++++++
.../templates/operator_configmap.yaml | 36 +++++++++++++++
deployments/gpu-operator/values.yaml | 2 +
3 files changed, 83 insertions(+)
create mode 100644 deployments/gpu-operator/templates/operator_configmap.yaml
diff --git a/deployments/gpu-operator/templates/operator.yaml b/deployments/gpu-operator/templates/operator.yaml
index 50983b20..1dfd9dbc 100644
--- a/deployments/gpu-operator/templates/operator.yaml
+++ b/deployments/gpu-operator/templates/operator.yaml
@@ -50,6 +50,45 @@ spec:
- name: host-os-release
mountPath: "/host-etc/os-release"
readOnly: true
+
+ {{- if eq .Values.operator.include_assets "include_assets" }}
+ {{- range $path, $_ := .Files.Glob "assets/gpu-feature-discovery/*" }}
+ - name: assets
+ mountPath: {{ printf "/opt/gpu-operator/gpu-feature-discovery/%s" (base $path) }}
+ subPath: {{ printf "gfd_%s" (base $path) }}
+ {{- end }}
+
+ {{- range $path, $_ := .Files.Glob "assets/state-container-toolkit/*" }}
+ - name: assets
+ mountPath: {{ printf "/opt/gpu-operator/state-container-toolkit/%s" (base $path) }}
+ subPath: {{ printf "state_container_toolkit_%s" (base $path) }}
+ {{- end }}
+
+ {{- range $path, $_ := .Files.Glob "assets/state-device-plugin/*" }}
+ - name: assets
+ mountPath: {{ printf "/opt/gpu-operator/state-device-plugin/%s" (base $path) }}
+ subPath: {{ printf "state_device_%s" (base $path) }}
+ {{- end }}
+
+ {{- range $path, $_ := .Files.Glob "assets/state-device-plugin-validation/*" }}
+ - name: assets
+ mountPath: {{ printf "/opt/gpu-operator/state-device-plugin-validation/%s" (base $path) }}
+ subPath: {{ printf "state_device_validation_%s" (base $path) }}
+ {{- end }}
+
+ {{- range $path, $_ := .Files.Glob "assets/state-driver/*" }}
+ - name: assets
+ mountPath: {{ printf "/opt/gpu-operator/state-driver/%s" (base $path) }}
+ subPath: {{ printf "state_driver_%s" (base $path) }}
+ {{- end }}
+
+ {{- range $path, $_ := .Files.Glob "assets/state-monitoring/*" }}
+ - name: assets
+ mountPath: {{ printf "/opt/gpu-operator/state-monitoring/%s" (base $path) }}
+ subPath: {{ printf "state_monitor_%s" (base $path) }}
+ {{- end }}
+ {{- end }}
+
readinessProbe:
exec:
command: ["stat", "/tmp/operator-sdk-ready"]
@@ -63,6 +102,12 @@ spec:
- name: host-os-release
hostPath:
path: "/etc/os-release"
+ {{- if eq .Values.operator.include_assets "include_assets" }}
+ - name: assets
+ configMap:
+ name: operator-configmap
+ {{- end }}
+
{{- with .Values.operator.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
diff --git a/deployments/gpu-operator/templates/operator_configmap.yaml b/deployments/gpu-operator/templates/operator_configmap.yaml
new file mode 100644
index 00000000..61f366e8
--- /dev/null
+++ b/deployments/gpu-operator/templates/operator_configmap.yaml
@@ -0,0 +1,36 @@
+{{- if eq .Values.operator.include_assets "include_assets" }}
+apiVersion: v1
+kind: ConfigMap
+metadata:
+ name: operator-configmap
+data:
+{{- range $path, $_ := .Files.Glob "assets/gpu-feature-discovery/*" }}
+{{ printf "gfd_%s" (base $path) | indent 2 }}: |-
+{{ $.Files.Get $path | indent 4 }}
+{{- end }}
+
+{{- range $path, $_ := .Files.Glob "assets/state-container-toolkit/*" }}
+{{ printf "state_container_toolkit_%s" (base $path) | indent 2 }}: |-
+{{ $.Files.Get $path | indent 4 }}
+{{- end }}
+
+{{- range $path, $_ := .Files.Glob "assets/state-device-plugin/*" }}
+{{ printf "state_device_%s" (base $path) | indent 2 }}: |-
+{{ $.Files.Get $path | indent 4 }}
+{{- end }}
+
+{{- range $path, $_ := .Files.Glob "assets/state-device-plugin-validation/*" }}
+{{ printf "state_device_validation_%s" (base $path) | indent 2 }}: |-
+{{ $.Files.Get $path | indent 4 }}
+{{- end }}
+
+{{- range $path, $_ := .Files.Glob "assets/state-driver/*" }}
+{{ printf "state_driver_%s" (base $path) | indent 2 }}: |-
+{{ $.Files.Get $path | indent 4 }}
+{{- end }}
+
+{{- range $path, $_ := .Files.Glob "assets/state-monitoring/*" }}
+{{ printf "state_monitor_%s" (base $path) | indent 2 }}: |-
+{{ $.Files.Get $path | indent 4 }}
+{{- end }}
+{{- end }}
diff --git a/deployments/gpu-operator/values.yaml b/deployments/gpu-operator/values.yaml
index 00d94195..8b43c59f 100644
--- a/deployments/gpu-operator/values.yaml
+++ b/deployments/gpu-operator/values.yaml
@@ -39,6 +39,8 @@ operator:
values: [""]
logging:
timeEncoding: epoch
+ # Set to "include_assets" to include assets/gpu-operator with the helm chart
+ include_assets: ""
driver:
repository: nvcr.io/nvidia
--
2.17.1

540
gpu/gpu-operator/files/enablement-support-on-starlingx-cloud-platform.patch Normal file
View File

@ -0,0 +1,540 @@
From 74c08e4ce69b80e8c5687d01c6bd1a4752233e20 Mon Sep 17 00:00:00 2001
From: Babak Sarashki <babak.sarashki@windriver.com>
Date: Sun, 7 Mar 2021 17:19:08 +0000
Subject: [PATCH 2/2] enablement: support on starlingx cloud platform
StarlingX is a cloud infrastructure software stack for edge.
It has an immutable file system, and system configruation. For
instance changes to set containerd runtime by the gpu-operator
will be overriden and must be avoided. The default_runtime is
to remain docker, therefore.
This commit enables gpu-operator on Starlingx (starlingx.io).
The changes to the gpu-operator include bundling modified assets
and a modified version of the nvidia-driver with the helm charts.
The modficiations to the assets include setting the runtimeClassName
on the gpu-operator pods that require nvidia-container-runtime and
host-mounting the kernel headers and build directory. The changes to
the nvidia-driver account for pre-installed kernel packages.
To load the operator on starlingx, define a runtimeclass with name
and handler set to nvidia; thereafter:
$ source /etc/platform/openrc
[...(keystone_admin)]$ system service-parameter-add \
platform container_runtime \
custom_container_runtime=nvidia:/path/to/nvidia-container-runtime
[...(keystone_admin)]$ system host-lock 1; system host-unlock 1
Signed-off-by: Babak Sarashki <babak.sarashki@windriver.com>
---
.../gpu-feature-discovery/0500_daemonset.yaml | 1 +
.../cuda-vector-add.yaml | 1 +
.../0400_device_plugin.yml | 1 +
assets/state-driver/0400_configmap.yaml | 327 +++++++++++++++++-
assets/state-driver/0500_daemonset.yaml | 39 ++-
assets/state-monitoring/0900_daemonset.yaml | 1 +
deployments/gpu-operator/values.yaml | 8 +-
7 files changed, 373 insertions(+), 5 deletions(-)
diff --git a/assets/gpu-feature-discovery/0500_daemonset.yaml b/assets/gpu-feature-discovery/0500_daemonset.yaml
index 9785dc93..1589e710 100644
--- a/assets/gpu-feature-discovery/0500_daemonset.yaml
+++ b/assets/gpu-feature-discovery/0500_daemonset.yaml
@@ -18,6 +18,7 @@ spec:
app.kubernetes.io/part-of: nvidia-gpu
spec:
serviceAccount: nvidia-gpu-feature-discovery
+ runtimeClassName: nvidia
containers:
- image: "FILLED BY THE OPERATOR"
name: gpu-feature-discovery
diff --git a/assets/state-device-plugin-validation/cuda-vector-add.yaml b/assets/state-device-plugin-validation/cuda-vector-add.yaml
index cfb547ad..8269adeb 100644
--- a/assets/state-device-plugin-validation/cuda-vector-add.yaml
+++ b/assets/state-device-plugin-validation/cuda-vector-add.yaml
@@ -12,6 +12,7 @@ spec:
effect: NoSchedule
readOnlyRootFilesystem: true
restartPolicy: OnFailure
+ runtimeClassName: nvidia
initContainers:
- name: device-plugin-validation-init
image: "FILLED BY THE OPERATOR"
diff --git a/assets/state-device-plugin/0400_device_plugin.yml b/assets/state-device-plugin/0400_device_plugin.yml
index a5cf7fae..84e9c534 100644
--- a/assets/state-device-plugin/0400_device_plugin.yml
+++ b/assets/state-device-plugin/0400_device_plugin.yml
@@ -30,6 +30,7 @@ spec:
operator: Exists
effect: NoSchedule
serviceAccount: nvidia-device-plugin
+ runtimeClassName: nvidia
initContainers:
- name: toolkit-validation
image: "FILLED BY THE OPERATOR"
diff --git a/assets/state-driver/0400_configmap.yaml b/assets/state-driver/0400_configmap.yaml
index 48e9f51e..561adc9f 100644
--- a/assets/state-driver/0400_configmap.yaml
+++ b/assets/state-driver/0400_configmap.yaml
@@ -4,7 +4,7 @@ metadata:
name: nvidia-driver
namespace: gpu-operator-resources
data:
- oci-nvidia-hook-json: |
+ oci-nvidia-hook-json: |
{
"version": "1.0.0",
"hook": {
@@ -20,3 +20,328 @@ data:
},
"stages": ["prestart"]
}
+ nvidia-driver-build-script: |
+ #! /bin/bash
+ # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
+ # Copyright (c) 2021 Wind River Systems, Inc. SPDX-License-Identifier:
+ # Apache-2.0.
+ # This script is from: https://gitlab.com/nvidia/container-images/driver.
+ # It is modified and included under configmap for platforms that require
+ # pre-installed packages. Such platforms have the option to modify the
+ # entrypoint in 0500_daemonset.yaml, or the nvidia-driver script here for
+ # further customizations.
+
+ set -eu
+
+ RUN_DIR=/run/nvidia
+ PID_FILE=${RUN_DIR}/${0##*/}.pid
+ DRIVER_VERSION=${DRIVER_VERSION:?"Missing driver version"}
+ KERNEL_UPDATE_HOOK=/run/kernel/postinst.d/update-nvidia-driver
+ KERNEL_VERSION="$(uname -r)"
+
+ # Default to 0 ; 1 is experimental and not supported
+ export IGNORE_PREEMPT_RT_PRESENCE=0
+
+ # Check if the kernel version requires a new precompiled driver packages.
+ _kernel_requires_package() {
+ local proc_mount_arg=""
+
+ echo "Checking NVIDIA driver packages..."
+ cd /usr/src/nvidia-${DRIVER_VERSION}/kernel
+
+ # When the kernel version is latest on host, this check fails and lead to recompilation, even when precompiled modules exist.
+ #if [ "${KERNEL_VERSION}" != "$(uname -r)" ]; then
+ #Not needed with pre-installed readonly headers, devel and modules
+ #proc_mount_arg="--proc-mount-point /lib/modules/${KERNEL_VERSION}/proc"
+ #fi
+ for pkg_name in $(ls -d -1 precompiled/** 2> /dev/null); do
+ is_match=$(../mkprecompiled --match ${pkg_name} ${proc_mount_arg})
+ if [ "${is_match}" == "kernel interface matches." ]; then
+ echo "Found NVIDIA driver package ${pkg_name##*/}"
+ return 1
+ fi
+ done
+ return 0
+ }
+
+ # Compile the kernel modules, optionally sign them, and generate a precompiled package for use by the nvidia-installer.
+ _create_driver_package() (
+ local pkg_name="nvidia-modules-${KERNEL_VERSION%%-*}${PACKAGE_TAG:+-${PACKAGE_TAG}}"
+ local nvidia_sign_args=""
+ local nvidia_modeset_sign_args=""
+ local nvidia_uvm_sign_args=""
+
+ trap "make -s -j 4 SYSSRC=/lib/modules/${KERNEL_VERSION}/build clean > /dev/null" EXIT
+
+ echo "Compiling NVIDIA driver kernel modules..."
+ cd /usr/src/nvidia-${DRIVER_VERSION}/kernel
+
+ export IGNORE_CC_MISMATCH=1
+ make -s -j 4 SYSSRC=/lib/modules/${KERNEL_VERSION}/build nv-linux.o nv-modeset-linux.o > /dev/null
+
+ echo "Relinking NVIDIA driver kernel modules..."
+ rm -f nvidia.ko nvidia-modeset.ko
+ ld -d -r -o nvidia.ko ./nv-linux.o ./nvidia/nv-kernel.o_binary
+ ld -d -r -o nvidia-modeset.ko ./nv-modeset-linux.o ./nvidia-modeset/nv-modeset-kernel.o_binary
+
+ if [ -n "${PRIVATE_KEY}" ]; then
+ echo "Signing NVIDIA driver kernel modules..."
+ donkey get ${PRIVATE_KEY} sh -c "PATH=${PATH}:/usr/src/kernels/$(uname -r)/scripts && \
+ sign-file sha512 \$DONKEY_FILE pubkey.x509 nvidia.ko nvidia.ko.sign && \
+ sign-file sha512 \$DONKEY_FILE pubkey.x509 nvidia-modeset.ko nvidia-modeset.ko.sign && \
+ sign-file sha512 \$DONKEY_FILE pubkey.x509 nvidia-uvm.ko"
+ nvidia_sign_args="--linked-module nvidia.ko --signed-module nvidia.ko.sign"
+ nvidia_modeset_sign_args="--linked-module nvidia-modeset.ko --signed-module nvidia-modeset.ko.sign"
+ nvidia_uvm_sign_args="--signed"
+ fi
+
+ echo "Building NVIDIA driver package ${pkg_name}..."
+ ../mkprecompiled --pack ${pkg_name} --description ${KERNEL_VERSION} \
+ --driver-version ${DRIVER_VERSION} \
+ --kernel-interface nv-linux.o \
+ --linked-module-name nvidia.ko \
+ --core-object-name nvidia/nv-kernel.o_binary \
+ ${nvidia_sign_args} \
+ --target-directory . \
+ --kernel-interface nv-modeset-linux.o \
+ --linked-module-name nvidia-modeset.ko \
+ --core-object-name nvidia-modeset/nv-modeset-kernel.o_binary \
+ ${nvidia_modeset_sign_args} \
+ --target-directory . \
+ --kernel-module nvidia-uvm.ko \
+ ${nvidia_uvm_sign_args} \
+ --target-directory .
+ mkdir -p precompiled
+ mv ${pkg_name} precompiled
+ )
+
+ # Load the kernel modules and start persistenced.
+ _load_driver() {
+ echo "Loading IPMI kernel module..."
+ modprobe ipmi_msghandler
+
+ echo "Loading NVIDIA driver kernel modules..."
+ modprobe -a nvidia nvidia-uvm nvidia-modeset
+
+ echo "Starting NVIDIA persistence daemon..."
+ nvidia-persistenced --persistence-mode
+ }
+
+ # Stop persistenced and unload the kernel modules if they are currently loaded.
+ _unload_driver() {
+ local rmmod_args=()
+ local nvidia_deps=0
+ local nvidia_refs=0
+ local nvidia_uvm_refs=0
+ local nvidia_modeset_refs=0
+
+ echo "Stopping NVIDIA persistence daemon..."
+ if [ -f /var/run/nvidia-persistenced/nvidia-persistenced.pid ]; then
+ local pid=$(< /var/run/nvidia-persistenced/nvidia-persistenced.pid)
+
+ kill -SIGTERM "${pid}"
+ for i in $(seq 1 10); do
+ kill -0 "${pid}" 2> /dev/null || break
+ sleep 0.1
+ done
+ if [ $i -eq 10 ]; then
+ echo "Could not stop NVIDIA persistence daemon" >&2
+ return 1
+ fi
+ fi
+
+ echo "Unloading NVIDIA driver kernel modules..."
+ if [ -f /sys/module/nvidia_modeset/refcnt ]; then
+ nvidia_modeset_refs=$(< /sys/module/nvidia_modeset/refcnt)
+ rmmod_args+=("nvidia-modeset")
+ ((++nvidia_deps))
+ fi
+ if [ -f /sys/module/nvidia_uvm/refcnt ]; then
+ nvidia_uvm_refs=$(< /sys/module/nvidia_uvm/refcnt)
+ rmmod_args+=("nvidia-uvm")
+ ((++nvidia_deps))
+ fi
+ if [ -f /sys/module/nvidia/refcnt ]; then
+ nvidia_refs=$(< /sys/module/nvidia/refcnt)
+ rmmod_args+=("nvidia")
+ fi
+ if [ ${nvidia_refs} -gt ${nvidia_deps} ] || [ ${nvidia_uvm_refs} -gt 0 ] || [ ${nvidia_modeset_refs} -gt 0 ]; then
+ echo "Could not unload NVIDIA driver kernel modules, driver is in use" >&2
+ return 1
+ fi
+
+ if [ ${#rmmod_args[@]} -gt 0 ]; then
+ rmmod ${rmmod_args[@]}
+ fi
+ return 0
+ }
+
+ # Link and install the kernel modules from a precompiled package using the nvidia-installer.
+ _install_driver() {
+ local install_args=()
+
+ echo "Installing NVIDIA driver kernel modules..."
+ cd /usr/src/nvidia-${DRIVER_VERSION}
+ rm -rf /lib/modules/${KERNEL_VERSION}/video
+
+ if [ "${ACCEPT_LICENSE}" = "yes" ]; then
+ install_args+=("--accept-license")
+ fi
+ nvidia-installer --kernel-module-only --no-drm --ui=none --no-nouveau-check ${install_args[@]+"${install_args[@]}"}
+ # May need to add no-cc-check for Rhel, otherwise it complains about cc missing in path
+ # /proc/version and lib/modules/KERNEL_VERSION/proc are different, by default installer looks at /proc/ so, added the proc-mount-point
+ # TODO: remove the -a flag. its not needed. in the new driver version, license-acceptance is implicit
+ #nvidia-installer --kernel-module-only --no-drm --ui=none --no-nouveau-check --no-cc-version-check --proc-mount-point /lib/modules/${KERNEL_VERSION}/proc ${install_args[@]+"${install_args[@]}"}
+ }
+
+ # Mount the driver rootfs into the run directory with the exception of sysfs.
+ _mount_rootfs() {
+ echo "Mounting NVIDIA driver rootfs..."
+ mount --make-runbindable /sys
+ mount --make-private /sys
+ mkdir -p ${RUN_DIR}/driver
+ mount --rbind / ${RUN_DIR}/driver
+ }
+
+ # Unmount the driver rootfs from the run directory.
+ _unmount_rootfs() {
+ echo "Unmounting NVIDIA driver rootfs..."
+ if findmnt -r -o TARGET | grep "${RUN_DIR}/driver" > /dev/null; then
+ umount -l -R ${RUN_DIR}/driver
+ fi
+ }
+
+ # Write a kernel postinst.d script to automatically precompile packages on kernel update (similar to DKMS).
+ _write_kernel_update_hook() {
+ if [ ! -d ${KERNEL_UPDATE_HOOK%/*} ]; then
+ return
+ fi
+
+ echo "Writing kernel update hook..."
+ cat > ${KERNEL_UPDATE_HOOK} <<'EOF'
+ #!/bin/bash
+
+ set -eu
+ trap 'echo "ERROR: Failed to update the NVIDIA driver" >&2; exit 0' ERR
+
+ NVIDIA_DRIVER_PID=$(< /run/nvidia/nvidia-driver.pid)
+
+ export "$(grep -z DRIVER_VERSION /proc/${NVIDIA_DRIVER_PID}/environ)"
+ nsenter -t "${NVIDIA_DRIVER_PID}" -m -- nvidia-driver update --kernel "$1"
+ EOF
+ chmod +x ${KERNEL_UPDATE_HOOK}
+ }
+
+ _shutdown() {
+ if _unload_driver; then
+ _unmount_rootfs
+ rm -f ${PID_FILE} ${KERNEL_UPDATE_HOOK}
+ return 0
+ fi
+ return 1
+ }
+
+ init() {
+ echo -e "\n========== NVIDIA Software Installer ==========\n"
+ echo -e "Starting installation of NVIDIA driver version ${DRIVER_VERSION} for Linux kernel version ${KERNEL_VERSION}\n"
+
+ exec 3> ${PID_FILE}
+ if ! flock -n 3; then
+ echo "An instance of the NVIDIA driver is already running, aborting"
+ exit 1
+ fi
+ echo $$ >&3
+
+ trap "echo 'Caught signal'; exit 1" HUP INT QUIT PIPE TERM
+ trap "_shutdown" EXIT
+
+ _unload_driver || exit 1
+ _unmount_rootfs
+
+ if _kernel_requires_package; then
+ _create_driver_package
+ fi
+
+ _install_driver
+ _load_driver
+ _mount_rootfs
+ _write_kernel_update_hook
+
+ echo "Done, now waiting for signal"
+ sleep infinity &
+ trap "echo 'Caught signal'; _shutdown && { kill $!; exit 0; }" HUP INT QUIT PIPE TERM
+ trap - EXIT
+ while true; do wait $! || continue; done
+ exit 0
+ }
+
+ update() {
+ exec 3>&2
+ if exec 2> /dev/null 4< ${PID_FILE}; then
+ if ! flock -n 4 && read pid <&4 && kill -0 "${pid}"; then
+ exec > >(tee -a "/proc/${pid}/fd/1")
+ exec 2> >(tee -a "/proc/${pid}/fd/2" >&3)
+ else
+ exec 2>&3
+ fi
+ exec 4>&-
+ fi
+ exec 3>&-
+
+ echo -e "\n========== NVIDIA Software Updater ==========\n"
+ echo -e "Starting update of NVIDIA driver version ${DRIVER_VERSION} for Linux kernel version ${KERNEL_VERSION}\n"
+
+ trap "echo 'Caught signal'; exit 1" HUP INT QUIT PIPE TERM
+
+ if _kernel_requires_package; then
+ _create_driver_package
+ fi
+
+ echo "Done"
+ exit 0
+ }
+
+ usage() {
+ cat >&2 <<EOF
+ Usage: $0 COMMAND [ARG...]
+
+ Commands:
+ init [-a | --accept-license]
+ update [-k | --kernel VERSION] [-s | --sign KEYID] [-t | --tag TAG]
+ EOF
+ exit 1
+ }
+
+ if [ $# -eq 0 ]; then
+ usage
+ fi
+ command=$1; shift
+ case "${command}" in
+ init) options=$(getopt -l accept-license -o a -- "$@") ;;
+ update) options=$(getopt -l kernel:,sign:,tag: -o k:s:t: -- "$@") ;;
+ *) usage ;;
+ esac
+ if [ $? -ne 0 ]; then
+ usage
+ fi
+ eval set -- "${options}"
+
+ ACCEPT_LICENSE=""
+ KERNEL_VERSION=$(uname -r)
+ PRIVATE_KEY=""
+ PACKAGE_TAG=""
+
+ for opt in ${options}; do
+ case "$opt" in
+ -a | --accept-license) ACCEPT_LICENSE="yes"; shift 1 ;;
+ -k | --kernel) KERNEL_VERSION=$2; shift 2 ;;
+ -s | --sign) PRIVATE_KEY=$2; shift 2 ;;
+ -t | --tag) PACKAGE_TAG=$2; shift 2 ;;
+ --) shift; break ;;
+ esac
+ done
+ if [ $# -ne 0 ]; then
+ usage
+ fi
+
+ $command
diff --git a/assets/state-driver/0500_daemonset.yaml b/assets/state-driver/0500_daemonset.yaml
index 3a2dc06b..7a1d8a17 100644
--- a/assets/state-driver/0500_daemonset.yaml
+++ b/assets/state-driver/0500_daemonset.yaml
@@ -32,8 +32,19 @@ spec:
- image: "FILLED BY THE OPERATOR"
imagePullPolicy: Always
name: nvidia-driver-ctr
- command: ["nvidia-driver"]
- args: ["init"]
+ command: ["/bin/bash"]
+ args:
+ - "-c"
+ - "--"
+ - >
+ cat /usr/local/bin/nvidia-driver.22 > /usr/local/bin/nvidia-driver &&
+ chmod 755 /usr/local/bin/nvidia-driver &&
+ mkdir -p /usr/src/kernels &&
+ tar -C /usr/src/host-kernels/ -c $(uname -r) -f - | tar -C /usr/src/kernels/ -xf - &&
+ rm -rf /lib/modules/ && mkdir -p /lib/modules/ &&
+ tar -C /lib/host-modules/ -c $(uname -r) -f - | tar -C /lib/modules/ -xf - &&
+ ln -rfs /usr/lib64/libelf.so.1 /usr/lib/libelf.so &&
+ /usr/local/bin/nvidia-driver init
securityContext:
privileged: true
seLinuxOptions:
@@ -44,10 +55,23 @@ spec:
mountPropagation: Bidirectional
- name: config
mountPath: /etc/containers/oci/hooks.d
+ subPath: oci-nvidia-hook-json
+ - name: config
+ mountPath: /usr/local/bin/nvidia-driver.22
+ subPath: nvidia-driver-build-script
- name: var-log
mountPath: /var/log
- name: dev-log
mountPath: /dev/log
+ - name: host-modules
+ mountPath: /lib/host-modules
+ readOnly: true
+ - name: host-include
+ mountPath: /usr/include
+ readOnly: true
+ - name: host-kernel-devel
+ mountPath: /usr/src/host-kernels
+ readOnly: true
volumes:
- name: run-nvidia
hostPath:
@@ -58,11 +82,22 @@ spec:
- name: dev-log
hostPath:
path: /dev/log
+ - name: host-modules
+ hostPath:
+ path: /lib/modules
+ - name: host-kernel-devel
+ hostPath:
+ path: /usr/src/kernels/
+ - name: host-include
+ hostPath:
+ path: /usr/include
- name: config
configMap:
name: nvidia-driver
items:
- key: oci-nvidia-hook-json
path: oci-nvidia-hook.json
+ - key: nvidia-driver-build-script
+ path: nvidia-driver-build-script
nodeSelector:
nvidia.com/gpu.present: "true"
diff --git a/assets/state-monitoring/0900_daemonset.yaml b/assets/state-monitoring/0900_daemonset.yaml
index 38c4d63a..aebb4297 100644
--- a/assets/state-monitoring/0900_daemonset.yaml
+++ b/assets/state-monitoring/0900_daemonset.yaml
@@ -31,6 +31,7 @@ spec:
effect: NoSchedule
serviceAccount: nvidia-dcgm-exporter
serviceAccountName: nvidia-dcgm-exporter
+ runtimeClassName: nvidia
initContainers:
- name: toolkit-validation
image: "FILLED BY THE OPERATOR"
diff --git a/deployments/gpu-operator/values.yaml b/deployments/gpu-operator/values.yaml
index 8b43c59f..17662729 100644
--- a/deployments/gpu-operator/values.yaml
+++ b/deployments/gpu-operator/values.yaml
@@ -15,6 +15,10 @@ operator:
#version: 1.5.2
imagePullPolicy: IfNotPresent
imagePullSecrets: []
+ # We cannot default to containerd because the operator modifies containerd
+ # configuration by adding itself to it, either as the default runtime or a
+ # runtimeclass, and restarts the service thereafter.
+ # defaultRuntime: containerd
defaultRuntime: docker
validator:
image: cuda-sample
@@ -40,7 +44,7 @@ operator:
logging:
timeEncoding: epoch
# Set to "include_assets" to include assets/gpu-operator with the helm chart
- include_assets: ""
+ include_assets: "include_assets"
driver:
repository: nvcr.io/nvidia
@@ -73,7 +77,7 @@ driver:
toolkit:
repository: nvcr.io/nvidia/k8s
image: container-toolkit
- version: 1.4.5-ubuntu18.04
+ version: 1.4.5-ubi8
imagePullPolicy: IfNotPresent
imagePullSecrets: []
env: []
--
2.17.1