前言
国内网络环境特殊，在线安装比较麻烦，K3S采用离线安装方式进行部署。
安装整体思路是：
- 安装GPU驱动
- 安装CUDA工具
- 安装nvidia容器运行时
- 安装K3S
- 设置K3S使用GPU
基础环境
采用All In One方式（其实只有一张GPU卡）部署。
| 参数 | 内容 |
|---|---|
| 系统 | openEuler 22.03 (LTS-SP3) |
| CPU | 8 |
| 内存 | 64G |
| 系统盘 | 500G |
| GPU | V100-32G |
GPU采用直通的方式，vGPU应该也差不多。
安装GPU驱动
下载驱动
先去官网(https://www.nvidia.cn/geforce/drivers/)下载驱动
选择对应型号操作系统为Linux 64-bit搜索后下载即可
像我V100的下载链接就是https://cn.download.nvidia.com/tesla/560.35.03/NVIDIA-Linux-x86_64-560.35.03.run
wget https://cn.download.nvidia.com/tesla/560.35.03/NVIDIA-Linux-x86_64-560.35.03.run -O NVIDIA-Linux.run
安装构建工具和依赖
yum install gcc make kernel-devel vulkan-loader -y
运行安装
根据指示安装完即可。
bash NVIDIA-Linux.run
测试是否安装成功
nvidia-smi
运行后应该会输出显卡和驱动信息，输出则表示安装成功
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 560.35.03 Driver Version: 560.35.03 CUDA Version: 12.6 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla V100-PCIE-32GB Off | 00000000:00:06.0 Off | 0 |
| N/A 41C P0 39W / 250W | 310MiB / 32768MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
安装CUDA工具
下载CUDA工具
前往官网（https://developer.nvidia.com/cuda-toolkit-archive），选择要下载的版本进行下载。
这里我选择的是12.4.0版本（PyTorch支持的版本），访问后操作系统选择Linux、架构选择x86_64、发行选择RHEL、版本选择9、安装方式选择runfile(local)。
根据安装指引运行
wget https://developer.download.nvidia.com/compute/cuda/12.4.0/local_installers/cuda_12.4.0_550.54.14_linux.run
sudo sh cuda_12.4.0_550.54.14_linux.run
校验CUDA
nvcc -V
运行后会返回CUDA版本信息
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:19:38_PST_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0
安装cuDNN
前往官网（https://developer.nvidia.com/rdp/cudnn-archive），选择要下载的版本进行下载。
访问后需要登录，根据指引下载至服务器即可，这个下载地址有时效性，需要尽快下载
下载完后运行解压安装即可
xz -d cudnn-linux-x86_64-8.9.7.29_cuda12-archive.tar.xz
tar -xvf cudnn-linux-x86_64-8.9.7.29_cuda12-archive.tar
cd cudnn-linux-x86_64-8.9.7.29_cuda12-archive/
/usr/bin/cp -r include/* /usr/local/cuda-12.4/include
/usr/bin/cp -r lib/* /usr/local/cuda-12.4/lib64
安装nvidia容器运行时
为了使容器内可以使用显卡，需要安装nvidia容器运行时，这一步网上经常是搜索的是ubuntu的教材，这里按官网步骤安装。
安装YUM源
curl -s -L https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo | \
sudo tee /etc/yum.repos.d/nvidia-container-toolkit.repo
启动YUM源
sudo dnf-config-manager --enable nvidia-container-toolkit-experimental
安装容器工具包
sudo dnf install -y nvidia-container-toolkit
安装完即可，在后续步骤会使用。
安装K3S
因为网络原因在线安装K3S不太稳定，这里使用离线安装。
下载镜像
mkdir -p /var/lib/rancher/k3s/agent/images/ /usr/local/bin/
wget https://github.com/k3s-io/k3s/releases/download/v1.31.5%2Bk3s1/k3s -O /usr/local/bin/k3s
chmod a+x /usr/local/bin/k3s
wget https://github.com/k3s-io/k3s/releases/download/v1.31.5%2Bk3s1/k3s-airgap-images-amd64.tar -O /var/lib/rancher/k3s/agent/images/k3s-airgap-images-amd64.tar
这里下载很慢可以使用类似镜像加速站点（https://gh-proxy.com）进行加速下载。
下载安装脚本
wget https://get.k3s.io -O install.sh
如果下载不了可以使用下面命令直接保存安装脚本
cat > install.sh <<EOF
#!/bin/sh
set -e
set -o noglob
# Usage:
# curl ... | ENV_VAR=... sh -
# or
# ENV_VAR=... ./install.sh
#
# Example:
# Installing a server without traefik:
# curl ... | INSTALL_K3S_EXEC="--disable=traefik" sh -
# Installing an agent to point at a server:
# curl ... | K3S_TOKEN=xxx K3S_URL=https://server-url:6443 sh -
#
# Environment variables:
# - K3S_*
# Environment variables which begin with K3S_ will be preserved for the
# systemd service to use. Setting K3S_URL without explicitly setting
# a systemd exec command will default the command to "agent", and we
# enforce that K3S_TOKEN is also set.
#
# - INSTALL_K3S_SKIP_DOWNLOAD
# If set to true will not download k3s hash or binary.
#
# - INSTALL_K3S_FORCE_RESTART
# If set to true will always restart the K3s service
#
# - INSTALL_K3S_SYMLINK
# If set to 'skip' will not create symlinks, 'force' will overwrite,
# default will symlink if command does not exist in path.
#
# - INSTALL_K3S_SKIP_ENABLE
# If set to true will not enable or start k3s service.
#
# - INSTALL_K3S_SKIP_START
# If set to true will not start k3s service.
#
# - INSTALL_K3S_VERSION
# Version of k3s to download from github. Will attempt to download from the
# stable channel if not specified.
#
# - INSTALL_K3S_COMMIT
# Commit of k3s to download from temporary cloud storage.
# * (for developer & QA use)
#
# - INSTALL_K3S_PR
# PR build of k3s to download from Github Artifacts.
# * (for developer & QA use)
#
# - INSTALL_K3S_BIN_DIR
# Directory to install k3s binary, links, and uninstall script to, or use
# /usr/local/bin as the default
#
# - INSTALL_K3S_BIN_DIR_READ_ONLY
# If set to true will not write files to INSTALL_K3S_BIN_DIR, forces
# setting INSTALL_K3S_SKIP_DOWNLOAD=true
#
# - INSTALL_K3S_SYSTEMD_DIR
# Directory to install systemd service and environment files to, or use
# /etc/systemd/system as the default
#
# - INSTALL_K3S_EXEC or script arguments
# Command with flags to use for launching k3s in the systemd service, if
# the command is not specified will default to "agent" if K3S_URL is set
# or "server" if not. The final systemd command resolves to a combination
# of EXEC and script args ($@).
#
# The following commands result in the same behavior:
# curl ... | INSTALL_K3S_EXEC="--disable=traefik" sh -s -
# curl ... | INSTALL_K3S_EXEC="server --disable=traefik" sh -s -
# curl ... | INSTALL_K3S_EXEC="server" sh -s - --disable=traefik
# curl ... | sh -s - server --disable=traefik
# curl ... | sh -s - --disable=traefik
#
# - INSTALL_K3S_NAME
# Name of systemd service to create, will default from the k3s exec command
# if not specified. If specified the name will be prefixed with 'k3s-'.
#
# - INSTALL_K3S_TYPE
# Type of systemd service to create, will default from the k3s exec command
# if not specified.
#
# - INSTALL_K3S_SELINUX_WARN
# If set to true will continue if k3s-selinux policy is not found.
#
# - INSTALL_K3S_SKIP_SELINUX_RPM
# If set to true will skip automatic installation of the k3s RPM.
#
# - INSTALL_K3S_CHANNEL_URL
# Channel URL for fetching k3s download URL.
# Defaults to 'https://update.k3s.io/v1-release/channels'.
#
# - INSTALL_K3S_CHANNEL
# Channel to use for fetching k3s download URL.
# Defaults to 'stable'.
GITHUB_URL=${GITHUB_URL:-https://github.com/k3s-io/k3s/releases}
GITHUB_PR_URL=""
STORAGE_URL=https://k3s-ci-builds.s3.amazonaws.com
DOWNLOADER=
# --- helper functions for logs ---
info()
{
echo '[INFO] ' "$@"
}
warn()
{
echo '[WARN] ' "$@" >&2
}
fatal()
{
echo '[ERROR] ' "$@" >&2
exit 1
}
# --- fatal if no systemd or openrc ---
verify_system() {
if [ -x /sbin/openrc-run ]; then
HAS_OPENRC=true
return
fi
if [ -x /bin/systemctl ] || type systemctl > /dev/null 2>&1; then
HAS_SYSTEMD=true
return
fi
fatal 'Can not find systemd or openrc to use as a process supervisor for k3s'
}
# --- add quotes to command arguments ---
quote() {
for arg in "$@"; do
printf '%s\n' "$arg" | sed "s/'/'\\\\''/g;1s/^/'/;\$s/\$/'/"
done
}
# --- add indentation and trailing slash to quoted args ---
quote_indent() {
printf ' \\\n'
for arg in "$@"; do
printf '\t%s \\\n' "$(quote "$arg")"
done
}
# --- escape most punctuation characters, except quotes, forward slash, and space ---
escape() {
printf '%s' "$@" | sed -e 's/\([][!#$%&()*;<=>?\_`{|}]\)/\\\1/g;'
}
# --- escape double quotes ---
escape_dq() {
printf '%s' "$@" | sed -e 's/"/\\"/g'
}
# --- ensures $K3S_URL is empty or begins with https://, exiting fatally otherwise ---
verify_k3s_url() {
case "${K3S_URL}" in
"")
;;
https://*)
;;
*)
fatal "Only https:// URLs are supported for K3S_URL (have ${K3S_URL})"
;;
esac
}
# --- define needed environment variables ---
setup_env() {
# --- use command args if passed or create default ---
case "$1" in
# --- if we only have flags discover if command should be server or agent ---
(-*|"")
if [ -z "${K3S_URL}" ]; then
CMD_K3S=server
else
if [ -z "${K3S_TOKEN}" ] && [ -z "${K3S_TOKEN_FILE}" ]; then
fatal "Defaulted k3s exec command to 'agent' because K3S_URL is defined, but K3S_TOKEN or K3S_TOKEN_FILE is not defined."
fi
CMD_K3S=agent
fi
;;
# --- command is provided ---
(*)
CMD_K3S=$1
shift
;;
esac
verify_k3s_url
CMD_K3S_EXEC="${CMD_K3S}$(quote_indent "$@")"
# --- use systemd name if defined or create default ---
if [ -n "${INSTALL_K3S_NAME}" ]; then
SYSTEM_NAME=k3s-${INSTALL_K3S_NAME}
else
if [ "${CMD_K3S}" = server ]; then
SYSTEM_NAME=k3s
else
SYSTEM_NAME=k3s-${CMD_K3S}
fi
fi
# --- check for invalid characters in system name ---
valid_chars=$(printf '%s' "${SYSTEM_NAME}" | sed -e 's/[][!#$%&()*;<=>?\_`{|}/[:space:]]/^/g;' )
if [ "${SYSTEM_NAME}" != "${valid_chars}" ]; then
invalid_chars=$(printf '%s' "${valid_chars}" | sed -e 's/[^^]/ /g')
fatal "Invalid characters for system name:
${SYSTEM_NAME}
${invalid_chars}"
fi
# --- use sudo if we are not already root ---
SUDO=sudo
if [ $(id -u) -eq 0 ]; then
SUDO=
fi
# --- use systemd type if defined or create default ---
if [ -n "${INSTALL_K3S_TYPE}" ]; then
SYSTEMD_TYPE=${INSTALL_K3S_TYPE}
else
SYSTEMD_TYPE=notify
fi
# --- use binary install directory if defined or create default ---
if [ -n "${INSTALL_K3S_BIN_DIR}" ]; then
BIN_DIR=${INSTALL_K3S_BIN_DIR}
else
# --- use /usr/local/bin if root can write to it, otherwise use /opt/bin if it exists
BIN_DIR=/usr/local/bin
if ! $SUDO sh -c "touch ${BIN_DIR}/k3s-ro-test && rm -rf ${BIN_DIR}/k3s-ro-test"; then
if [ -d /opt/bin ]; then
BIN_DIR=/opt/bin
fi
fi
fi
# --- use systemd directory if defined or create default ---
if [ -n "${INSTALL_K3S_SYSTEMD_DIR}" ]; then
SYSTEMD_DIR="${INSTALL_K3S_SYSTEMD_DIR}"
else
SYSTEMD_DIR=/etc/systemd/system
fi
# --- set related files from system name ---
SERVICE_K3S=${SYSTEM_NAME}.service
UNINSTALL_K3S_SH=${UNINSTALL_K3S_SH:-${BIN_DIR}/${SYSTEM_NAME}-uninstall.sh}
KILLALL_K3S_SH=${KILLALL_K3S_SH:-${BIN_DIR}/k3s-killall.sh}
# --- use service or environment location depending on systemd/openrc ---
if [ "${HAS_SYSTEMD}" = true ]; then
FILE_K3S_SERVICE=${SYSTEMD_DIR}/${SERVICE_K3S}
FILE_K3S_ENV=${SYSTEMD_DIR}/${SERVICE_K3S}.env
elif [ "${HAS_OPENRC}" = true ]; then
$SUDO mkdir -p /etc/rancher/k3s
FILE_K3S_SERVICE=/etc/init.d/${SYSTEM_NAME}
FILE_K3S_ENV=/etc/rancher/k3s/${SYSTEM_NAME}.env
fi
# --- get hash of config & exec for currently installed k3s ---
PRE_INSTALL_HASHES=$(get_installed_hashes)
# --- if bin directory is read only skip download ---
if [ "${INSTALL_K3S_BIN_DIR_READ_ONLY}" = true ]; then
INSTALL_K3S_SKIP_DOWNLOAD=true
fi
# --- setup channel values
INSTALL_K3S_CHANNEL_URL=${INSTALL_K3S_CHANNEL_URL:-'https://update.k3s.io/v1-release/channels'}
INSTALL_K3S_CHANNEL=${INSTALL_K3S_CHANNEL:-'stable'}
}
# --- check if skip download environment variable set ---
can_skip_download_binary() {
if [ "${INSTALL_K3S_SKIP_DOWNLOAD}" != true ] && [ "${INSTALL_K3S_SKIP_DOWNLOAD}" != binary ]; then
return 1
fi
}
can_skip_download_selinux() {
if [ "${INSTALL_K3S_SKIP_DOWNLOAD}" != true ] && [ "${INSTALL_K3S_SKIP_DOWNLOAD}" != selinux ]; then
return 1
fi
}
# --- verify an executable k3s binary is installed ---
verify_k3s_is_executable() {
if [ ! -x ${BIN_DIR}/k3s ]; then
fatal "Executable k3s binary not found at ${BIN_DIR}/k3s"
fi
}
# --- set arch and suffix, fatal if architecture not supported ---
setup_verify_arch() {
if [ -z "$ARCH" ]; then
ARCH=$(uname -m)
fi
case $ARCH in
amd64)
ARCH=amd64
SUFFIX=
;;
x86_64)
ARCH=amd64
SUFFIX=
;;
arm64)
ARCH=arm64
SUFFIX=-${ARCH}
;;
s390x)
ARCH=s390x
SUFFIX=-${ARCH}
;;
aarch64)
ARCH=arm64
SUFFIX=-${ARCH}
;;
arm*)
ARCH=arm
SUFFIX=-${ARCH}hf
;;
*)
fatal "Unsupported architecture $ARCH"
esac
}
# --- verify existence of network downloader executable ---
verify_downloader() {
# Return failure if it doesn't exist or is no executable
[ -x "$(command -v $1)" ] || return 1
# Set verified executable as our downloader program and return success
DOWNLOADER=$1
return 0
}
# --- create temporary directory and cleanup when done ---
setup_tmp() {
TMP_DIR=$(mktemp -d -t k3s-install.XXXXXXXXXX)
TMP_HASH=${TMP_DIR}/k3s.hash
TMP_ZIP=${TMP_DIR}/k3s.zip
TMP_BIN=${TMP_DIR}/k3s.bin
cleanup() {
code=$?
set +e
trap - EXIT
rm -rf ${TMP_DIR}
exit $code
}
trap cleanup INT EXIT
}
# --- use desired k3s version if defined or find version from channel ---
get_release_version() {
if [ -n "${INSTALL_K3S_PR}" ]; then
VERSION_K3S="PR ${INSTALL_K3S_PR}"
get_pr_artifact_url
elif [ -n "${INSTALL_K3S_COMMIT}" ]; then
VERSION_K3S="commit ${INSTALL_K3S_COMMIT}"
elif [ -n "${INSTALL_K3S_VERSION}" ]; then
VERSION_K3S=${INSTALL_K3S_VERSION}
else
info "Finding release for channel ${INSTALL_K3S_CHANNEL}"
version_url="${INSTALL_K3S_CHANNEL_URL}/${INSTALL_K3S_CHANNEL}"
case $DOWNLOADER in
curl)
VERSION_K3S=$(curl -w '%{url_effective}' -L -s -S ${version_url} -o /dev/null | sed -e 's|.*/||')
;;
wget)
VERSION_K3S=$(wget -SqO /dev/null ${version_url} 2>&1 | grep -i Location | sed -e 's|.*/||')
;;
*)
fatal "Incorrect downloader executable '$DOWNLOADER'"
;;
esac
fi
info "Using ${VERSION_K3S} as release"
}
# --- get k3s-selinux version ---
get_k3s_selinux_version() {
available_version="k3s-selinux-1.2-2.${rpm_target}.noarch.rpm"
info "Finding available k3s-selinux versions"
# run verify_downloader in case it binary installation was skipped
verify_downloader curl || verify_downloader wget || fatal 'Can not find curl or wget for downloading files'
case $DOWNLOADER in
curl)
DOWNLOADER_OPTS="-s"
;;
wget)
DOWNLOADER_OPTS="-q -O -"
;;
*)
fatal "Incorrect downloader executable '$DOWNLOADER'"
;;
esac
for i in {1..3}; do
set +e
if [ "${rpm_channel}" = "testing" ]; then
version=$(timeout 5 ${DOWNLOADER} ${DOWNLOADER_OPTS} https://api.github.com/repos/k3s-io/k3s-selinux/releases | grep browser_download_url | awk '{ print $2 }' | grep -oE "[^\/]+${rpm_target}\.noarch\.rpm" | head -n 1)
else
version=$(timeout 5 ${DOWNLOADER} ${DOWNLOADER_OPTS} https://api.github.com/repos/k3s-io/k3s-selinux/releases/latest | grep browser_download_url | awk '{ print $2 }' | grep -oE "[^\/]+${rpm_target}\.noarch\.rpm")
fi
set -e
if [ "${version}" != "" ]; then
break
fi
sleep 1
done
if [ "${version}" == "" ]; then
warn "Failed to get available versions of k3s-selinux..defaulting to ${available_version}"
return
fi
available_version=${version}
}
# --- download from github url ---
download() {
[ $# -eq 2 ] || fatal 'download needs exactly 2 arguments'
# Disable exit-on-error so we can do custom error messages on failure
set +e
# Default to a failure status
status=1
case $DOWNLOADER in
curl)
curl -o $1 -sfL $2
status=$?
;;
wget)
wget -qO $1 $2
status=$?
;;
*)
# Enable exit-on-error for fatal to execute
set -e
fatal "Incorrect executable '$DOWNLOADER'"
;;
esac
# Re-enable exit-on-error
set -e
# Abort if download command failed
[ $status -eq 0 ] || fatal 'Download failed'
}
# --- download hash from github url ---
download_hash() {
if [ -n "${INSTALL_K3S_PR}" ]; then
info "Downloading hash ${GITHUB_PR_URL}"
curl -s -o ${TMP_ZIP} -H "Authorization: Bearer $GITHUB_TOKEN" -L ${GITHUB_PR_URL}
unzip -p ${TMP_ZIP} k3s.sha256sum > ${TMP_HASH}
else
if [ -n "${INSTALL_K3S_COMMIT}" ]; then
HASH_URL=${STORAGE_URL}/k3s${SUFFIX}-${INSTALL_K3S_COMMIT}.sha256sum
else
HASH_URL=${GITHUB_URL}/download/${VERSION_K3S}/sha256sum-${ARCH}.txt
fi
info "Downloading hash ${HASH_URL}"
download ${TMP_HASH} ${HASH_URL}
fi
HASH_EXPECTED=$(grep " k3s${SUFFIX}$" ${TMP_HASH})
HASH_EXPECTED=${HASH_EXPECTED%%[[:blank:]]*}
}
# --- check hash against installed version ---
installed_hash_matches() {
if [ -x ${BIN_DIR}/k3s ]; then
HASH_INSTALLED=$(sha256sum ${BIN_DIR}/k3s)
HASH_INSTALLED=${HASH_INSTALLED%%[[:blank:]]*}
if [ "${HASH_EXPECTED}" = "${HASH_INSTALLED}" ]; then
return
fi
fi
return 1
}
# Use the GitHub API to identify the artifact associated with a given PR
get_pr_artifact_url() {
github_api_url=https://api.github.com/repos/k3s-io/k3s
# Check if jq is installed
if ! [ -x "$(command -v jq)" ]; then
fatal "Installing PR builds requires jq"
fi
# Check if unzip is installed
if ! [ -x "$(command -v unzip)" ]; then
fatal "Installing PR builds requires unzip"
fi
if [ -z "${GITHUB_TOKEN}" ]; then
fatal "Installing PR builds requires GITHUB_TOKEN with k3s-io/k3s repo permissions"
fi
# GET request to the GitHub API to retrieve the latest commit SHA from the pull request
set +e
commit_id=$(curl -f -s -H "Authorization: Bearer ${GITHUB_TOKEN}" "${github_api_url}/pulls/${INSTALL_K3S_PR}" | jq -r '.head.sha')
set -e
if [ -z "${commit_id}" ]; then
fatal "Installing PR builds requires GITHUB_TOKEN with k3s-io/k3s repo permissions"
fi
# GET request to the GitHub API to retrieve the Build workflow associated with the commit
run_id=$(curl -s -H "Authorization: Bearer ${GITHUB_TOKEN}" "${github_api_url}/commits/${commit_id}/check-runs?check_name=build%20%2F%20Build" | jq -r '[.check_runs | sort_by(.id) | .[].details_url | split("/")[7]] | last')
# Extract the artifact ID for the "k3s" artifact
GITHUB_PR_URL=$(curl -s -H "Authorization: Bearer ${GITHUB_TOKEN}" "${github_api_url}/actions/runs/${run_id}/artifacts" | jq -r '.artifacts[] | select(.name == "k3s") | .archive_download_url')
}
# --- download binary from github url ---
download_binary() {
if [ -n "${INSTALL_K3S_PR}" ]; then
# Since Binary and Hash are zipped together, check if TMP_ZIP already exists
if ! [ -f ${TMP_ZIP} ]; then
info "Downloading K3s artifact ${GITHUB_PR_URL}"
curl -s -f -o ${TMP_ZIP} -H "Authorization: Bearer $GITHUB_TOKEN" -L ${GITHUB_PR_URL}
fi
# extract k3s binary from zip
unzip -p ${TMP_ZIP} k3s > ${TMP_BIN}
return
elif [ -n "${INSTALL_K3S_COMMIT}" ]; then
BIN_URL=${STORAGE_URL}/k3s${SUFFIX}-${INSTALL_K3S_COMMIT}
else
BIN_URL=${GITHUB_URL}/download/${VERSION_K3S}/k3s${SUFFIX}
fi
info "Downloading binary ${BIN_URL}"
download ${TMP_BIN} ${BIN_URL}
}
# --- verify downloaded binary hash ---
verify_binary() {
info "Verifying binary download"
HASH_BIN=$(sha256sum ${TMP_BIN})
HASH_BIN=${HASH_BIN%%[[:blank:]]*}
if [ "${HASH_EXPECTED}" != "${HASH_BIN}" ]; then
fatal "Download sha256 does not match ${HASH_EXPECTED}, got ${HASH_BIN}"
fi
}
# --- setup permissions and move binary to system directory ---
setup_binary() {
chmod 755 ${TMP_BIN}
info "Installing k3s to ${BIN_DIR}/k3s"
$SUDO chown root:root ${TMP_BIN}
$SUDO mv -f ${TMP_BIN} ${BIN_DIR}/k3s
}
# --- setup selinux policy ---
setup_selinux() {
case ${INSTALL_K3S_CHANNEL} in
*testing)
rpm_channel=testing
;;
*latest)
rpm_channel=latest
;;
*)
rpm_channel=stable
;;
esac
rpm_site="rpm.rancher.io"
if [ "${rpm_channel}" = "testing" ]; then
rpm_site="rpm-testing.rancher.io"
fi
[ -r /etc/os-release ] && . /etc/os-release
if [ `expr "${ID_LIKE}" : ".*suse.*"` != 0 ]; then
rpm_target=sle
rpm_site_infix=microos
package_installer=zypper
if [ "${ID_LIKE:-}" = suse ] && ( [ "${VARIANT_ID:-}" = sle-micro ] || [ "${ID:-}" = sle-micro ] ); then
rpm_target=sle
rpm_site_infix=slemicro
package_installer=zypper
fi
elif [ "${ID_LIKE:-}" = coreos ] || [ "${VARIANT_ID:-}" = coreos ] || [ "${VARIANT_ID:-}" = "iot" ] || \
{ { [ "${ID:-}" = fedora ] || [ "${ID_LIKE:-}" = fedora ]; } && [ -n "${OSTREE_VERSION:-}" ]; }; then
rpm_target=coreos
rpm_site_infix=coreos
package_installer=rpm-ostree
elif [ "${VERSION_ID%%.*}" = "7" ] || ( [ "${ID:-}" = amzn ] && [ "${VERSION_ID%%.*}" = "2" ] ); then
rpm_target=el7
rpm_site_infix=centos/7
package_installer=yum
elif [ "${VERSION_ID%%.*}" = "8" ] || [ "${VERSION_ID%%.*}" = "V10" ] || [ "${VERSION_ID%%.*}" -gt "36" ]; then
rpm_target=el8
rpm_site_infix=centos/8
package_installer=yum
else
rpm_target=el9
rpm_site_infix=centos/9
package_installer=yum
fi
if [ "${package_installer}" = "rpm-ostree" ] && [ -x /bin/yum ]; then
package_installer=yum
fi
if [ "${package_installer}" = "yum" ] && [ -x /usr/bin/dnf ]; then
package_installer=dnf
fi
policy_hint="please install:
${package_installer} install -y container-selinux
${package_installer} install -y https://${rpm_site}/k3s/${rpm_channel}/common/${rpm_site_infix}/noarch/${available_version}
"
if [ "$INSTALL_K3S_SKIP_SELINUX_RPM" = true ] || can_skip_download_selinux || [ ! -d /usr/share/selinux ]; then
info "Skipping installation of SELinux RPM"
return
fi
get_k3s_selinux_version
install_selinux_rpm ${rpm_site} ${rpm_channel} ${rpm_target} ${rpm_site_infix}
policy_error=fatal
if [ "$INSTALL_K3S_SELINUX_WARN" = true ] || [ "${ID_LIKE:-}" = coreos ] ||
[ "${VARIANT_ID:-}" = coreos ] || [ "${VARIANT_ID:-}" = iot ]; then
policy_error=warn
fi
if ! $SUDO chcon -u system_u -r object_r -t container_runtime_exec_t ${BIN_DIR}/k3s >/dev/null 2>&1; then
if $SUDO grep '^\s*SELINUX=enforcing' /etc/selinux/config >/dev/null 2>&1; then
$policy_error "Failed to apply container_runtime_exec_t to ${BIN_DIR}/k3s, ${policy_hint}"
fi
elif [ ! -f /usr/share/selinux/packages/k3s.pp ]; then
if [ -x /usr/sbin/transactional-update ] || [ "${ID_LIKE:-}" = coreos ] || \
{ { [ "${ID:-}" = fedora ] || [ "${ID_LIKE:-}" = fedora ]; } && [ -n "${OSTREE_VERSION:-}" ]; }; then
warn "Please reboot your machine to activate the changes and avoid data loss."
else
$policy_error "Failed to find the k3s-selinux policy, ${policy_hint}"
fi
fi
}
install_selinux_rpm() {
if [ -r /etc/redhat-release ] || [ -r /etc/centos-release ] || [ -r /etc/oracle-release ] ||
[ -r /etc/fedora-release ] || [ -r /etc/system-release ] || [ "${ID_LIKE%%[ ]*}" = "suse" ]; then
repodir=/etc/yum.repos.d
if [ -d /etc/zypp/repos.d ]; then
repodir=/etc/zypp/repos.d
fi
set +o noglob
$SUDO rm -f ${repodir}/rancher-k3s-common*.repo
set -o noglob
if [ -r /etc/redhat-release ] && [ "${3}" = "el7" ]; then
$SUDO yum install -y yum-utils
$SUDO yum-config-manager --enable rhel-7-server-extras-rpms
fi
$SUDO tee ${repodir}/rancher-k3s-common.repo >/dev/null << EOF
[rancher-k3s-common-${2}]
name=Rancher K3s Common (${2})
baseurl=https://${1}/k3s/${2}/common/${4}/noarch
enabled=1
gpgcheck=1
repo_gpgcheck=0
gpgkey=https://${1}/public.key
EOF
case ${3} in
sle)
rpm_installer="zypper --gpg-auto-import-keys"
if [ "${TRANSACTIONAL_UPDATE=false}" != "true" ] && [ -x /usr/sbin/transactional-update ]; then
transactional_update_run="transactional-update --no-selfupdate -d run"
rpm_installer="transactional-update --no-selfupdate -d run ${rpm_installer}"
: "${INSTALL_K3S_SKIP_START:=true}"
fi
# create the /var/lib/rpm-state in SLE systems to fix the prein selinux macro
$SUDO ${transactional_update_run} mkdir -p /var/lib/rpm-state
;;
coreos)
rpm_installer="rpm-ostree --idempotent"
# rpm_install_extra_args="--apply-live"
: "${INSTALL_K3S_SKIP_START:=true}"
;;
*)
rpm_installer="yum"
;;
esac
if [ "${rpm_installer}" = "yum" ] && [ -x /usr/bin/dnf ]; then
rpm_installer=dnf
fi
if rpm -q --quiet k3s-selinux; then
# remove k3s-selinux module before upgrade to allow container-selinux to upgrade safely
if check_available_upgrades container-selinux ${3} && check_available_upgrades k3s-selinux ${3}; then
MODULE_PRIORITY=$($SUDO semodule --list=full | grep k3s | cut -f1 -d" ")
if [ -n "${MODULE_PRIORITY}" ]; then
$SUDO semodule -X $MODULE_PRIORITY -r k3s || true
fi
fi
fi
# shellcheck disable=SC2086
$SUDO ${rpm_installer} install -y "k3s-selinux"
fi
return
}
check_available_upgrades() {
set +e
case ${2} in
sle)
available_upgrades=$($SUDO zypper -q -t -s 11 se -s -u --type package $1 | tail -n 1 | grep -v "No matching" | awk '{print $3}')
;;
coreos)
# currently rpm-ostree does not support search functionality https://github.com/coreos/rpm-ostree/issues/1877
;;
*)
available_upgrades=$($SUDO yum -q --refresh list $1 --upgrades | tail -n 1 | awk '{print $2}')
;;
esac
set -e
if [ -n "${available_upgrades}" ]; then
return 0
fi
return 1
}
# --- download and verify k3s ---
download_and_verify() {
if can_skip_download_binary; then
info 'Skipping k3s download and verify'
verify_k3s_is_executable
return
fi
setup_verify_arch
verify_downloader curl || verify_downloader wget || fatal 'Can not find curl or wget for downloading files'
setup_tmp
get_release_version
download_hash
if installed_hash_matches; then
info 'Skipping binary downloaded, installed k3s matches hash'
return
fi
download_binary
verify_binary
setup_binary
}
# --- add additional utility links ---
create_symlinks() {
[ "${INSTALL_K3S_BIN_DIR_READ_ONLY}" = true ] && return
[ "${INSTALL_K3S_SYMLINK}" = skip ] && return
for cmd in kubectl crictl ctr; do
if [ ! -e ${BIN_DIR}/${cmd} ] || [ "${INSTALL_K3S_SYMLINK}" = force ]; then
which_cmd=$(command -v ${cmd} 2>/dev/null || true)
if [ -z "${which_cmd}" ] || [ "${INSTALL_K3S_SYMLINK}" = force ]; then
info "Creating ${BIN_DIR}/${cmd} symlink to k3s"
$SUDO ln -sf k3s ${BIN_DIR}/${cmd}
else
info "Skipping ${BIN_DIR}/${cmd} symlink to k3s, command exists in PATH at ${which_cmd}"
fi
else
info "Skipping ${BIN_DIR}/${cmd} symlink to k3s, already exists"
fi
done
}
# --- create killall script ---
create_killall() {
[ "${INSTALL_K3S_BIN_DIR_READ_ONLY}" = true ] && return
info "Creating killall script ${KILLALL_K3S_SH}"
$SUDO tee ${KILLALL_K3S_SH} >/dev/null << \EOF
#!/bin/sh
[ $(id -u) -eq 0 ] || exec sudo --preserve-env=K3S_DATA_DIR $0 $@
K3S_DATA_DIR=${K3S_DATA_DIR:-/var/lib/rancher/k3s}
for bin in ${K3S_DATA_DIR}/data/**/bin/; do
[ -d $bin ] && export PATH=$PATH:$bin:$bin/aux
done
set -x
for service in /etc/systemd/system/k3s*.service; do
[ -s $service ] && systemctl stop $(basename $service)
done
for service in /etc/init.d/k3s*; do
[ -x $service ] && $service stop
done
pschildren() {
ps -e -o ppid= -o pid= | \
sed -e 's/^\s*//g; s/\s\s*/\t/g;' | \
grep -w "^$1" | \
cut -f2
}
pstree() {
for pid in $@; do
echo $pid
for child in $(pschildren $pid); do
pstree $child
done
done
}
killtree() {
kill -9 $(
{ set +x; } 2>/dev/null;
pstree $@;
set -x;
) 2>/dev/null
}
remove_interfaces() {
# Delete network interface(s) that match 'master cni0'
ip link show 2>/dev/null | grep 'master cni0' | while read ignore iface ignore; do
iface=${iface%%@*}
[ -z "$iface" ] || ip link delete $iface
done
# Delete cni related interfaces
ip link delete cni0
ip link delete flannel.1
ip link delete flannel-v6.1
ip link delete kube-ipvs0
ip link delete flannel-wg
ip link delete flannel-wg-v6
# Restart tailscale
if [ -n "$(command -v tailscale)" ]; then
tailscale set --advertise-routes=
fi
}
getshims() {
ps -e -o pid= -o args= | sed -e 's/^ *//; s/\s\s*/\t/;' | grep -w "${K3S_DATA_DIR}"'/data/[^/]*/bin/containerd-shim' | cut -f1
}
killtree $({ set +x; } 2>/dev/null; getshims; set -x)
do_unmount_and_remove() {
set +x
while read -r _ path _; do
case "$path" in $1*) echo "$path" ;; esac
done < /proc/self/mounts | sort -r | xargs -r -t -n 1 sh -c 'umount -f "$0" && rm -rf "$0"'
set -x
}
do_unmount_and_remove '/run/k3s'
do_unmount_and_remove '/var/lib/kubelet/pods'
do_unmount_and_remove '/var/lib/kubelet/plugins'
do_unmount_and_remove '/run/netns/cni-'
# Remove CNI namespaces
ip netns show 2>/dev/null | grep cni- | xargs -r -t -n 1 ip netns delete
remove_interfaces
rm -rf /var/lib/cni/
iptables-save | grep -v KUBE- | grep -v CNI- | grep -iv flannel | iptables-restore
ip6tables-save | grep -v KUBE- | grep -v CNI- | grep -iv flannel | ip6tables-restore
EOF
$SUDO chmod 755 ${KILLALL_K3S_SH}
$SUDO chown root:root ${KILLALL_K3S_SH}
}
# --- create uninstall script ---
create_uninstall() {
[ "${INSTALL_K3S_BIN_DIR_READ_ONLY}" = true ] && return
info "Creating uninstall script ${UNINSTALL_K3S_SH}"
$SUDO tee ${UNINSTALL_K3S_SH} >/dev/null << EOF
#!/bin/sh
set -x
[ \$(id -u) -eq 0 ] || exec sudo --preserve-env=K3S_DATA_DIR \$0 \$@
K3S_DATA_DIR=\${K3S_DATA_DIR:-/var/lib/rancher/k3s}
${KILLALL_K3S_SH}
if command -v systemctl; then
systemctl disable ${SYSTEM_NAME}
systemctl reset-failed ${SYSTEM_NAME}
systemctl daemon-reload
fi
if command -v rc-update; then
rc-update delete ${SYSTEM_NAME} default
fi
rm -f ${FILE_K3S_SERVICE}
rm -f ${FILE_K3S_ENV}
remove_uninstall() {
rm -f ${UNINSTALL_K3S_SH}
}
trap remove_uninstall EXIT
if (ls ${SYSTEMD_DIR}/k3s*.service || ls /etc/init.d/k3s*) >/dev/null 2>&1; then
set +x; echo 'Additional k3s services installed, skipping uninstall of k3s'; set -x
exit
fi
for cmd in kubectl crictl ctr; do
if [ -L ${BIN_DIR}/\$cmd ]; then
rm -f ${BIN_DIR}/\$cmd
fi
done
clean_mounted_directory() {
if ! grep -q " \$1" /proc/mounts; then
rm -rf "\$1"
return 0
fi
for path in "\$1"/*; do
if [ -d "\$path" ]; then
if grep -q " \$path" /proc/mounts; then
clean_mounted_directory "\$path"
else
rm -rf "\$path"
fi
else
rm "\$path"
fi
done
}
rm -rf /etc/rancher/k3s
rm -rf /run/k3s
rm -rf /run/flannel
clean_mounted_directory \${K3S_DATA_DIR}
rm -rf /var/lib/kubelet
rm -f ${BIN_DIR}/k3s
rm -f ${KILLALL_K3S_SH}
if type yum >/dev/null 2>&1; then
yum remove -y k3s-selinux
rm -f /etc/yum.repos.d/rancher-k3s-common*.repo
elif type rpm-ostree >/dev/null 2>&1; then
rpm-ostree uninstall k3s-selinux
rm -f /etc/yum.repos.d/rancher-k3s-common*.repo
elif type zypper >/dev/null 2>&1; then
uninstall_cmd="zypper remove -y k3s-selinux"
if [ "\${TRANSACTIONAL_UPDATE=false}" != "true" ] && [ -x /usr/sbin/transactional-update ]; then
uninstall_cmd="transactional-update --no-selfupdate -d run \$uninstall_cmd"
fi
$SUDO \$uninstall_cmd
rm -f /etc/zypp/repos.d/rancher-k3s-common*.repo
fi
EOF
$SUDO chmod 755 ${UNINSTALL_K3S_SH}
$SUDO chown root:root ${UNINSTALL_K3S_SH}
}
# --- disable current service if loaded --
systemd_disable() {
$SUDO systemctl disable ${SYSTEM_NAME} >/dev/null 2>&1 || true
$SUDO rm -f /etc/systemd/system/${SERVICE_K3S} || true
$SUDO rm -f /etc/systemd/system/${SERVICE_K3S}.env || true
}
# --- capture current env and create file containing k3s_ variables ---
create_env_file() {
info "env: Creating environment file ${FILE_K3S_ENV}"
$SUDO touch ${FILE_K3S_ENV}
$SUDO chmod 0600 ${FILE_K3S_ENV}
sh -c export | while read x v; do echo $v; done | grep -E '^(K3S|CONTAINERD)_' | $SUDO tee ${FILE_K3S_ENV} >/dev/null
sh -c export | while read x v; do echo $v; done | grep -Ei '^(NO|HTTP|HTTPS)_PROXY' | $SUDO tee -a ${FILE_K3S_ENV} >/dev/null
}
# --- write systemd service file ---
create_systemd_service_file() {
info "systemd: Creating service file ${FILE_K3S_SERVICE}"
$SUDO tee ${FILE_K3S_SERVICE} >/dev/null << EOF
[Unit]
Description=Lightweight Kubernetes
Documentation=https://k3s.io
Wants=network-online.target
After=network-online.target
[Install]
WantedBy=multi-user.target
[Service]
Type=${SYSTEMD_TYPE}
EnvironmentFile=-/etc/default/%N
EnvironmentFile=-/etc/sysconfig/%N
EnvironmentFile=-${FILE_K3S_ENV}
KillMode=process
Delegate=yes
# Having non-zero Limit*s causes performance problems due to accounting overhead
# in the kernel. We recommend using cgroups to do container-local accounting.
LimitNOFILE=1048576
LimitNPROC=infinity
LimitCORE=infinity
TasksMax=infinity
TimeoutStartSec=0
Restart=always
RestartSec=5s
ExecStartPre=/bin/sh -xc '! /usr/bin/systemctl is-enabled --quiet nm-cloud-setup.service 2>/dev/null'
ExecStartPre=-/sbin/modprobe br_netfilter
ExecStartPre=-/sbin/modprobe overlay
ExecStart=${BIN_DIR}/k3s \\
${CMD_K3S_EXEC}
EOF
}
# --- write openrc service file ---
create_openrc_service_file() {
LOG_FILE=/var/log/${SYSTEM_NAME}.log
info "openrc: Creating service file ${FILE_K3S_SERVICE}"
$SUDO tee ${FILE_K3S_SERVICE} >/dev/null << EOF
#!/sbin/openrc-run
depend() {
after network-online
want cgroups
}
start_pre() {
rm -f /tmp/k3s.*
}
supervisor=supervise-daemon
name=${SYSTEM_NAME}
command="${BIN_DIR}/k3s"
command_args="$(escape_dq "${CMD_K3S_EXEC}")
>>${LOG_FILE} 2>&1"
output_log=${LOG_FILE}
error_log=${LOG_FILE}
pidfile="/var/run/${SYSTEM_NAME}.pid"
respawn_delay=5
respawn_max=0
set -o allexport
if [ -f /etc/environment ]; then . /etc/environment; fi
if [ -f ${FILE_K3S_ENV} ]; then . ${FILE_K3S_ENV}; fi
set +o allexport
EOF
$SUDO chmod 0755 ${FILE_K3S_SERVICE}
$SUDO tee /etc/logrotate.d/${SYSTEM_NAME} >/dev/null << EOF
${LOG_FILE} {
missingok
notifempty
copytruncate
}
EOF
}
# --- write systemd or openrc service file ---
create_service_file() {
[ "${HAS_SYSTEMD}" = true ] && create_systemd_service_file && restore_systemd_service_file_context
[ "${HAS_OPENRC}" = true ] && create_openrc_service_file
return 0
}
restore_systemd_service_file_context() {
$SUDO restorecon -R -i ${FILE_K3S_SERVICE} 2>/dev/null || true
$SUDO restorecon -R -i ${FILE_K3S_ENV} 2>/dev/null || true
}
# --- get hashes of the current k3s bin and service files
get_installed_hashes() {
$SUDO sha256sum ${BIN_DIR}/k3s ${FILE_K3S_SERVICE} ${FILE_K3S_ENV} 2>&1 || true
}
# --- enable and start systemd service ---
systemd_enable() {
info "systemd: Enabling ${SYSTEM_NAME} unit"
$SUDO systemctl enable ${FILE_K3S_SERVICE} >/dev/null
$SUDO systemctl daemon-reload >/dev/null
}
systemd_start() {
info "systemd: Starting ${SYSTEM_NAME}"
$SUDO systemctl restart ${SYSTEM_NAME}
}
# --- enable and start openrc service ---
openrc_enable() {
info "openrc: Enabling ${SYSTEM_NAME} service for default runlevel"
$SUDO rc-update add ${SYSTEM_NAME} default >/dev/null
}
openrc_start() {
info "openrc: Starting ${SYSTEM_NAME}"
$SUDO ${FILE_K3S_SERVICE} restart
}
has_working_xtables() {
if $SUDO sh -c "command -v \"$1-save\"" 1> /dev/null && $SUDO sh -c "command -v \"$1-restore\"" 1> /dev/null; then
if $SUDO $1-save 2>/dev/null | grep -q '^-A CNI-HOSTPORT-MASQ -j MASQUERADE$'; then
warn "Host $1-save/$1-restore tools are incompatible with existing rules"
else
return 0
fi
else
info "Host $1-save/$1-restore tools not found"
fi
return 1
}
# --- startup systemd or openrc service ---
service_enable_and_start() {
if [ -f "/proc/cgroups" ] && [ "$(grep memory /proc/cgroups | while read -r n n n enabled; do echo $enabled; done)" -eq 0 ];
then
info 'Failed to find memory cgroup, you may need to add "cgroup_memory=1 cgroup_enable=memory" to your linux cmdline (/boot/cmdline.txt on a Raspberry Pi)'
fi
[ "${INSTALL_K3S_SKIP_ENABLE}" = true ] && return
[ "${HAS_SYSTEMD}" = true ] && systemd_enable
[ "${HAS_OPENRC}" = true ] && openrc_enable
[ "${INSTALL_K3S_SKIP_START}" = true ] && return
POST_INSTALL_HASHES=$(get_installed_hashes)
if [ "${PRE_INSTALL_HASHES}" = "${POST_INSTALL_HASHES}" ] && [ "${INSTALL_K3S_FORCE_RESTART}" != true ]; then
info 'No change detected so skipping service start'
return
fi
for XTABLES in iptables ip6tables; do
if has_working_xtables ${XTABLES}; then
$SUDO ${XTABLES}-save 2>/dev/null | grep -v KUBE- | grep -iv flannel | $SUDO ${XTABLES}-restore
fi
done
[ "${HAS_SYSTEMD}" = true ] && systemd_start
[ "${HAS_OPENRC}" = true ] && openrc_start
return 0
}
# --- re-evaluate args to include env command ---
eval set -- $(escape "${INSTALL_K3S_EXEC}") $(quote "$@")
# --- run the install process --
{
verify_system
setup_env "$@"
download_and_verify
setup_selinux
create_symlinks
create_killall
create_uninstall
systemd_disable
create_env_file
create_service_file
service_enable_and_start
}
EOF
安装K3S
export INSTALL_K3S_EXEC="--system-default-registry registry.cn-hangzhou.aliyuncs.com --write-kubeconfig ~/.kube/config --disable=traefik --embedded-registry --service-node-port-range=443-32767"
# 设置使用国内源
export INSTALL_K3S_MIRROR=cn
# 设置强制下载
export INSTALL_K3S_SYMLINK=force
# 设置镜像url
export INSTALL_K3S_MIRROR_URL=${INSTALL_K3S_MIRROR_URL:-'rancher-mirror.rancher.cn'}
export INSTALL_K3S_SKIP_DOWNLOAD=true
bash install.sh
配置镜像源
安装完成后还需要配置镜像源，不然大概率使用中会出现ImagePullErr的错误
cat >/etc/rancher/k3s/registries.yaml <<EOF
mirrors:
gcr.io:
quay.io:
ghcr.io:
registry.k8s.io:
docker.io:
endpoint:
- "https://docker.1panel.live"
- "https://docker.1ms.run"
EOF
systemctl restart k3s
设置K3S使用GPU
安装完K3S默认是不能使用GPU的，因为用的是默认运行时，默认运行时没有调用GPU的能力。
网上文档大多是讲修改Docker运行时或者完整k8s的操作，经过踩坑后我整理的使用步骤如下：
修改配置文件更改k3s默认运行时为nvidia容器运行时
nvidia-ctk runtime configure --runtime=containerd --set-as-default --config /var/lib/rancher/k3s/agent/etc/containerd/config.toml
配置持久化
k3s每次重启都会生成新的配置，需要将配置保存成tmpl文件才能自动加载
cp /var/lib/rancher/k3s/agent/etc/containerd/config.toml /var/lib/rancher/k3s/agent/etc/containerd/config.toml.tmpl
重启使配置生效
systemctl restart k3s
安装nvidia容器驱动
kubectl apply -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.14.5/nvidia-device-plugin.yml
运行后检查是否安装成功
kubectl get pods -n kube-system | grep nvidia-device-plugin
安装成功后应该会出现类似如下的输出
nvidia-device-plugin-daemonset-mbptx 1/1 Running 0 78m
这时候需要检查设备插件日志，确认是否有报错:
kubectl logs -n kube-system nvidia-device-plugin-daemonset-mbptx
看到这样的字段表示设备成功注册
I0220 11:14:12.852130 1 main.go:356] Retrieving plugins.
I0220 11:14:12.957036 1 server.go:195] Starting GRPC server for 'nvidia.com/gpu'
I0220 11:14:12.995786 1 server.go:139] Starting to serve 'nvidia.com/gpu' on /var/lib/kubelet/device-plugins/nvidia-gpu.sock
I0220 11:14:13.030801 1 server.go:146] Registered device plugin for 'nvidia.com/gpu' with Kubelet
测试K3S调用GPU
最后校验运行k3s使用GPU
cat > gpu-test.yaml <<EOF
apiVersion: v1
kind: Pod
metadata:
name: nvidia-smi-test
spec:
restartPolicy: OnFailure
containers:
- name: nvidia-smi
image: nvidia/cuda:12.1.0-base-ubuntu22.04
command: ["nvidia-smi"]
resources:
limits:
nvidia.com/gpu: 1
EOF
kubectl create -f gpu-test.yaml
kubectl logs pod/nvidia-smi-test
最后应该可以成功认到显卡输出
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 560.35.03 Driver Version: 560.35.03 CUDA Version: 12.6 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla V100-PCIE-32GB Off | 00000000:00:06.0 Off | 0 |
| N/A 41C P0 39W / 250W | 310MiB / 32768MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+
至此配置完成。