ai-workspace-infra/gitops
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(deepflow): split image pull script into arm64/x86

Browse Source 
- removed legacy `pull_save_scp_image.sh`
- added separate scripts for arm64 and x86
- updated `images.txt` and `deploy_deepflow_agent.sh`
- adjusted OpenResty config templates (`artifact.conf.j2`, `nginx.conf.j2`)
This commit is contained in:
Haitao Pan 2025-08-28 11:46:14 +08:00
parent 7961f103b6
commit ea2bbc6acd
7 changed files with 538 additions and 69 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
playbooks/roles/vhosts/OpenResty/templates/artifact.conf.j2
View File

@ -16,24 +16,22 @@ server {
{% set autoindex_paths = item.autoindex_paths | default(vhost_defaults.autoindex_paths) %}
# 目录浏览（打开 autoindex）—可列出整个 {{ item.root | default(vhost_defaults.root) }}
{% if '/' in autoindex_paths %}
location / {
{% if '/' in autoindex_paths %}
autoindex on;
autoindex_exact_size off;
autoindex_localtime on;
{% endif %}
add_header Accept-Ranges bytes;
try_files $uri $uri/ =404; # 保持原有 404 语义
}
{% for path in autoindex_paths %}
{% if path != '/' %}
{% elif %}
{% for path in autoindex_paths %}
location {{ path }} {
autoindex on;
autoindex_exact_size off;
autoindex_localtime on;
}
{% endif %}
{% endif %}
{% endfor %}
# 常见安装包直下读文件（大小写不敏感）

1
playbooks/roles/vhosts/OpenResty/templates/nginx.conf.j2
View File

@ -8,6 +8,7 @@ events {
http {
include mime.types;
default_type application/octet-stream;
lua_package_path "/usr/local/openresty/lualib/?.lua;;";
sendfile on;
keepalive_timeout 65;

6
scripts/deepflow/images.txt Normal file
View File

@ -0,0 +1,6 @@
dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/df-analyze:latest.515
dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/df-web-agent:latest.2986036
dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/df-web-composer:latest.2984869
dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/apientry:latest.268
hub.deepflow.yunshan.net/dev/alpine:latest
hub.deepflow.yunshan.net/dev/weaviate:1.30.0

32
scripts/deepflow/pull_save_scp_image.sh
View File

@ -1,32 +0,0 @@
#!/bin/bash
set -e
REMOTE_HOST="root@10.1.3.179"
if [ -z "$1" ]; then
echo "❌ 用法: $0 <image>"
echo "示例: $0 dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/df-web-ai:v6.6.18839"
exit 1
fi
IMAGE="$1"
# 提取镜像名和版本号
NAME_TAG="${IMAGE##*/}" # df-web-ai:v6.6.18839
NAME="${NAME_TAG%%:*}" # df-web-ai
TAG="${NAME_TAG##*:}" # v6.6.18839
FILE_NAME="${NAME}-${TAG//v/}.tar" # df-web-ai-6.6.18839.tar
echo "📦 镜像: $IMAGE"
echo "📁 导出文件名: $FILE_NAME"
echo "🚀 在远程拉取镜像..."
ssh $REMOTE_HOST docker pull "$IMAGE"
echo "💾 在远程保存镜像为 /tmp/$FILE_NAME..."
ssh $REMOTE_HOST "docker save $IMAGE > /tmp/$FILE_NAME"
echo "📥 拷贝镜像回本地 ~/Desktop..."
scp $REMOTE_HOST:/tmp/$FILE_NAME ~/Desktop
echo "✅ 完成！镜像保存于：~/Desktop/$FILE_NAME"

198
scripts/deepflow/pull_save_scp_image_arm64.sh Normal file
View File

@ -0,0 +1,198 @@
#!/bin/bash
# deepflow/pull_save_scp_image_arm64.sh
# 目标：即使远端是 x86，也只拉取/保存 arm64 变体，并强校验；支持批量与 --rm-remote 清理
set -euo pipefail
REMOTE_HOST="${REMOTE_HOST:-root@10.1.3.179}"
DEST_DIR="${DEST_DIR:-$HOME/Desktop}"
RM_REMOTE=0
usage() {
cat <<EOF
用法:
$0 <image1> [image2 ...] [--rm-remote]
$0 -f images.txt [--rm-remote]
说明:
- 支持批量处理:
• 多个参数: ./pull_save_scp_image_arm64.sh image1 image2 ...
• 文件清单: ./pull_save_scp_image_arm64.sh -f images.txt
(清单支持 # 注释与空行)
- 只拉 arm64 & save arm64:
• docker pull --platform=linux/arm64
• docker image inspect --format '{{.Architecture}}' 二次确认
• 以镜像ID保存，避免 tag→manifest list 在 x86 上回退到 amd64
- 保存后校验:
• 在远端解析 tar 的 manifest.json 和对应 config
• 逐个检查 "architecture":"arm64"，确保 tar 内确实是 arm64
- 可配置环境变量:
• REMOTE_HOST (默认 root@10.1.3.179)
• DEST_DIR (默认 ~/Desktop)
- 额外选项:
• --rm-remote 成功拷贝到本地后自动删除远端 /tmp/*.tar；
任一步失败也会自动清理远端临时文件，避免残留。
示例:
$0 dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/df-web-ai:v6.6.18839
$0 -f images.txt --rm-remote
EOF
}
# -------- 参数解析 --------
IMAGES=()
LIST_FILE=""
if [[ $# -eq 0 ]]; then usage; exit 1; fi
ARGS=()
while [[ $# -gt 0 ]]; do
case "$1" in
-h|--help) usage; exit 0 ;;
--rm-remote) RM_REMOTE=1; shift ;;
-f)
[[ $# -ge 2 ]] || { echo "❌ 缺少镜像清单文件"; exit 1; }
LIST_FILE="$2"; shift 2 ;;
*)
ARGS+=("$1"); shift ;;
esac
done
if [[ -n "$LIST_FILE" ]]; then
[[ -f "$LIST_FILE" ]] || { echo "❌ 文件不存在: $LIST_FILE"; exit 1; }
while IFS= read -r line; do
line="${line%%#*}"
line="$(echo -n "$line" | xargs || true)"
[[ -n "$line" ]] || continue
IMAGES+=("$line")
done < "$LIST_FILE"
fi
if [[ ${#ARGS[@]} -gt 0 ]]; then
IMAGES+=("${ARGS[@]}")
fi
[[ ${#IMAGES[@]} -gt 0 ]] || { echo "❌ 没有可处理的镜像"; exit 1; }
echo "🖥️ 远端: $REMOTE_HOST"
echo "💾 本地保存目录: $DEST_DIR"
echo "🧹 rm-remote: $([[ $RM_REMOTE -eq 1 ]] && echo ON || echo OFF)"
mkdir -p "$DEST_DIR"
# -------- 远端校验脚本内容（用 cat heredoc 赋值，兼容 macOS 老 bash） --------
REMOTE_VERIFY_PY="$(cat <<'PYCODE'
import sys, tarfile, json
tar_path = sys.argv[1]
with tarfile.open(tar_path, "r") as tf:
manifest = json.load(tf.extractfile("manifest.json"))
for item in manifest:
cfg = item.get("Config")
if not cfg:
print("NO_CONFIG_IN_MANIFEST", file=sys.stderr); sys.exit(2)
f = tf.extractfile(cfg)
if f is None:
print("CONFIG_NOT_FOUND", file=sys.stderr); sys.exit(3)
cfg_json = json.load(f)
arch = cfg_json.get("architecture")
if arch != "arm64":
print(f"BAD_ARCH:{arch}", file=sys.stderr); sys.exit(4)
print("OK")
PYCODE
)"
# 为了安全传输到远端，这里将 Python 内容做一次 printf %q 转义
#（避免某些 shell/ssh 环境下的字符解释问题）
escape_for_ssh() {
printf "%s" "$1" | python3 - <<'P'
import sys, shlex
data=sys.stdin.read()
print(shlex.quote(data))
P
}
REMOTE_VERIFY_PY_Q=$(escape_for_ssh "$REMOTE_VERIFY_PY")
# -------- 处理单个镜像 --------
process_image() {
local IMAGE="$1"
local NAME_TAG="${IMAGE##*/}" # e.g. weaviate:1.30.0
local NAME="${NAME_TAG%%:*}" # weaviate
local TAG="${NAME_TAG##*:}" # 1.30.0
if [[ "$NAME" == "$NAME_TAG" ]]; then TAG="latest"; fi
local FILE_NAME="${NAME}-${TAG}.arm64.tar"
local REMOTE_TAR="/tmp/${FILE_NAME}"
local DEST_PATH="${DEST_DIR}/${FILE_NAME}"
echo
echo "=============================="
echo "📦 镜像: $IMAGE"
echo "🎯 仅拉取平台: linux/arm64"
echo "📁 导出文件名: $FILE_NAME"
echo "=============================="
# 失败即清理远端临时文件
local CLEAN_ON_FAILURE=1
trap 'if [[ "${CLEAN_ON_FAILURE:-0}" -eq 1 ]]; then ssh -o BatchMode=yes "'"$REMOTE_HOST"'" "rm -f \"'"$REMOTE_TAR"'\"" || true; fi' RETURN
# 1) 强制拉 arm64
echo "🚀 远端拉取镜像..."
ssh -o BatchMode=yes "$REMOTE_HOST" "docker pull --platform=linux/arm64 \"$IMAGE\""
# 2) 获取 arm64 变体镜像ID
echo "🔎 提取 arm64 镜像ID..."
local IMAGE_ID
IMAGE_ID="$(ssh "$REMOTE_HOST" "
docker image inspect \"$IMAGE\" \
--format '{{.Id}} {{.Architecture}}' 2>/dev/null \
| awk '\$2==\"arm64\"{print \$1; exit}'
")"
if [[ -z "${IMAGE_ID:-}" ]]; then
echo "❌ 未找到 arm64 变体镜像ID，可能仓库不包含 arm64。"; return 12
fi
echo "✅ arm64 镜像ID: $IMAGE_ID"
# 3) 二次确认该 ID 的架构为 arm64
echo "🧪 inspect 架构确认..."
ssh "$REMOTE_HOST" "
arch=\$(docker image inspect --format '{{.Architecture}}' $IMAGE_ID | head -n1); \
if [[ \"\$arch\" != \"arm64\" ]]; then
echo '❌ 镜像ID架构校验失败: '\"\$arch\"; exit 13; fi; \
echo '✅ 架构: '\"\$arch\"
"
# 4) 以镜像ID保存
echo "💾 保存为: $REMOTE_TAR ..."
ssh "$REMOTE_HOST" "docker save $IMAGE_ID > \"$REMOTE_TAR\""
# 5) 解包校验 tar 内架构
echo "🧬 校验 tar 包内部 architecture..."
ssh "$REMOTE_HOST" "python3 -c $REMOTE_VERIFY_PY_Q \"$REMOTE_TAR\""
# 6) 拷回本地
echo "📥 拷贝到本地: $DEST_PATH ..."
scp "$REMOTE_HOST:$REMOTE_TAR" "$DEST_PATH"
# 7) 可选删除远端临时 tar；关闭失败清理 trap
if [[ $RM_REMOTE -eq 1 ]]; then
echo "🧹 删除远端临时文件: $REMOTE_TAR"
ssh "$REMOTE_HOST" "rm -f \"$REMOTE_TAR\""
else
echo "ℹ️ 远端临时文件保留: $REMOTE_TAR"
fi
CLEAN_ON_FAILURE=0
trap - RETURN
echo "✅ 完成: $DEST_PATH (arm64 only)"
}
# -------- 批量执行 --------
for img in "${IMAGES[@]}"; do
process_image "$img"
done
echo
echo "🎉 所有任务完成。"

196
scripts/deepflow/pull_save_scp_image_x86.sh Normal file
View File

@ -0,0 +1,196 @@
#!/bin/bash
# deepflow/pull_save_scp_image_amd64.sh
# 目标：仅拉取/保存 amd64 变体，并强校验；支持批量与 --rm-remote 清理
set -euo pipefail
REMOTE_HOST="${REMOTE_HOST:-root@10.1.3.179}"
DEST_DIR="${DEST_DIR:-$HOME/Desktop}"
RM_REMOTE=0
usage() {
cat <<EOF
用法:
$0 <image1> [image2 ...] [--rm-remote]
$0 -f images.txt [--rm-remote]
说明:
- 支持批量处理:
• 多个参数: ./pull_save_scp_image_amd64.sh image1 image2 ...
• 文件清单: ./pull_save_scp_image_amd64.sh -f images.txt
(清单支持 # 注释与空行)
- 只拉 amd64 & save amd64:
• docker pull --platform=linux/amd64
• docker image inspect --format '{{.Architecture}}' 二次确认
• 以镜像ID保存，避免 tag→manifest list 在异构主机上回退到其他架构
- 保存后校验:
• 在远端解析 tar 的 manifest.json 和对应 config
• 逐个检查 "architecture":"amd64"，确保 tar 内确实是 amd64
- 可配置环境变量:
• REMOTE_HOST (默认 root@10.1.3.179)
• DEST_DIR (默认 ~/Desktop)
- 额外选项:
• --rm-remote 成功拷贝到本地后自动删除远端 /tmp/*.tar；
任一步失败也会自动清理远端临时文件，避免残留。
示例:
$0 dfcloud-image-registry-vpc.cn-beijing.cr.aliyuncs.com/dev/df-web-ai:v6.6.18839
$0 -f images.txt --rm-remote
EOF
}
# -------- 参数解析 --------
IMAGES=()
LIST_FILE=""
if [[ $# -eq 0 ]]; then usage; exit 1; fi
ARGS=()
while [[ $# -gt 0 ]]; do
case "$1" in
-h|--help) usage; exit 0 ;;
--rm-remote) RM_REMOTE=1; shift ;;
-f)
[[ $# -ge 2 ]] || { echo "❌ 缺少镜像清单文件"; exit 1; }
LIST_FILE="$2"; shift 2 ;;
*)
ARGS+=("$1"); shift ;;
esac
done
if [[ -n "$LIST_FILE" ]]; then
[[ -f "$LIST_FILE" ]] || { echo "❌ 文件不存在: $LIST_FILE"; exit 1; }
while IFS= read -r line; do
line="${line%%#*}"
line="$(echo -n "$line" | xargs || true)"
[[ -n "$line" ]] || continue
IMAGES+=("$line")
done < "$LIST_FILE"
fi
if [[ ${#ARGS[@]} -gt 0 ]]; then
IMAGES+=("${ARGS[@]}")
fi
[[ ${#IMAGES[@]} -gt 0 ]] || { echo "❌ 没有可处理的镜像"; exit 1; }
echo "🖥️ 远端: $REMOTE_HOST"
echo "💾 本地保存目录: $DEST_DIR"
echo "🧹 rm-remote: $([[ $RM_REMOTE -eq 1 ]] && echo ON || echo OFF)"
mkdir -p "$DEST_DIR"
# -------- 远端校验脚本内容（用 cat heredoc 赋值，兼容 macOS 老 bash） --------
REMOTE_VERIFY_PY="$(cat <<'PYCODE'
import sys, tarfile, json
tar_path = sys.argv[1]
with tarfile.open(tar_path, "r") as tf:
manifest = json.load(tf.extractfile("manifest.json"))
for item in manifest:
cfg = item.get("Config")
if not cfg:
print("NO_CONFIG_IN_MANIFEST", file=sys.stderr); sys.exit(2)
f = tf.extractfile(cfg)
if f is None:
print("CONFIG_NOT_FOUND", file=sys.stderr); sys.exit(3)
cfg_json = json.load(f)
arch = cfg_json.get("architecture")
if arch != "amd64":
print(f"BAD_ARCH:{arch}", file=sys.stderr); sys.exit(4)
print("OK")
PYCODE
)"
escape_for_ssh() {
printf "%s" "$1" | python3 - <<'P'
import sys, shlex
data=sys.stdin.read()
print(shlex.quote(data))
P
}
REMOTE_VERIFY_PY_Q=$(escape_for_ssh "$REMOTE_VERIFY_PY")
# -------- 处理单个镜像 --------
process_image() {
local IMAGE="$1"
local NAME_TAG="${IMAGE##*/}" # e.g. weaviate:1.30.0
local NAME="${NAME_TAG%%:*}" # weaviate
local TAG="${NAME_TAG##*:}" # 1.30.0
if [[ "$NAME" == "$NAME_TAG" ]]; then TAG="latest"; fi
local FILE_NAME="${NAME}-${TAG}.amd64.tar"
local REMOTE_TAR="/tmp/${FILE_NAME}"
local DEST_PATH="${DEST_DIR}/${FILE_NAME}"
echo
echo "=============================="
echo "📦 镜像: $IMAGE"
echo "🎯 仅拉取平台: linux/amd64"
echo "📁 导出文件名: $FILE_NAME"
echo "=============================="
# 失败即清理远端临时文件
local CLEAN_ON_FAILURE=1
trap 'if [[ "${CLEAN_ON_FAILURE:-0}" -eq 1 ]]; then ssh -o BatchMode=yes "'"$REMOTE_HOST"'" "rm -f \"'"$REMOTE_TAR"'\"" || true; fi' RETURN
# 1) 强制拉 amd64
echo "🚀 远端拉取镜像..."
ssh -o BatchMode=yes "$REMOTE_HOST" "docker pull --platform=linux/amd64 \"$IMAGE\""
# 2) 获取 amd64 变体镜像ID
echo "🔎 提取 amd64 镜像ID..."
local IMAGE_ID
IMAGE_ID="$(ssh "$REMOTE_HOST" "
docker image inspect \"$IMAGE\" \
--format '{{.Id}} {{.Architecture}}' 2>/dev/null \
| awk '\$2==\"amd64\"{print \$1; exit}'
")"
if [[ -z "${IMAGE_ID:-}" ]]; then
echo "❌ 未找到 amd64 变体镜像ID，可能仓库不包含 amd64。"; return 12
fi
echo "✅ amd64 镜像ID: $IMAGE_ID"
# 3) 二次确认该 ID 的架构为 amd64
echo "🧪 inspect 架构确认..."
ssh "$REMOTE_HOST" "
arch=\$(docker image inspect --format '{{.Architecture}}' $IMAGE_ID | head -n1); \
if [[ \"\$arch\" != \"amd64\" ]]; then
echo '❌ 镜像ID架构校验失败: '\"\$arch\"; exit 13; fi; \
echo '✅ 架构: '\"\$arch\"
"
# 4) 以镜像ID保存
echo "💾 保存为: $REMOTE_TAR ..."
ssh "$REMOTE_HOST" "docker save $IMAGE_ID > \"$REMOTE_TAR\""
# 5) 解包校验 tar 内架构
echo "🧬 校验 tar 包内部 architecture..."
ssh "$REMOTE_HOST" "python3 -c $REMOTE_VERIFY_PY_Q \"$REMOTE_TAR\""
# 6) 拷回本地
echo "📥 拷贝到本地: $DEST_PATH ..."
scp "$REMOTE_HOST:$REMOTE_TAR" "$DEST_PATH"
# 7) 可选删除远端临时 tar；关闭失败清理 trap
if [[ $RM_REMOTE -eq 1 ]]; then
echo "🧹 删除远端临时文件: $REMOTE_TAR"
ssh "$REMOTE_HOST" "rm -f \"$REMOTE_TAR\""
else
echo "ℹ️ 远端临时文件保留: $REMOTE_TAR"
fi
CLEAN_ON_FAILURE=0
trap - RETURN
echo "✅ 完成: $DEST_PATH (amd64 only)"
}
# -------- 批量执行 --------
for img in "${IMAGES[@]}"; do
process_image "$img"
done
echo
echo "🎉 所有任务完成。"

164
scripts/deploy_deepflow_agent.sh
View File

@ -3,7 +3,7 @@
set -e
####################################
# 🌐 配置区
# <EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 配置区
####################################
IP_LIST="./ip.list"
@ -14,6 +14,7 @@ MAX_PARALLEL=5
CONTROLLER_IP=""
VTAP_GROUP_ID=""
LIMIT=""
SUDO_MODE="sudo" # 可选: sudo | sudo-i
SSH_OPTS="-o StrictHostKeyChecking=no -o ConnectTimeout=15"
@ -23,11 +24,11 @@ SUCCESS_FILE="success_hosts.txt"
> "$SUCCESS_FILE"
####################################
# 参数解析
# 参数解析（新增 --sudo-mode）
####################################
if [[ $# -eq 0 ]]; then
echo "用法: $0 {deploy|upgrade|verify} --controller <ip> --group <id> [--limit ip1,ip2]"
echo "用法: $0 {deploy|upgrade|verify} --controller <ip> --group <id> [--limit ip1,ip2] [--sudo-mode sudo|sudo-i]"
exit 1
fi
@ -48,6 +49,18 @@ while [[ $# -gt 0 ]]; do
LIMIT="$2"
shift 2
;;
--sudo-mode)
case "$2" in
sudo|sudo-i)
SUDO_MODE="$2"
shift 2
;;
*)
echo "❌ --sudo-mode 必须是 'sudo' 或 'sudo-i'"
exit 1
;;
esac
;;
*)
echo "未知参数: $1"
exit 1
@ -56,7 +69,7 @@ while [[ $# -gt 0 ]]; do
done
if [[ "$ACTION" != "deploy" && "$ACTION" != "upgrade" && "$ACTION" != "verify" ]]; then
echo "用法: $0 {deploy|upgrade|verify} --controller <ip> --group <id> [--limit ip1,ip2]"
echo "用法: $0 {deploy|upgrade|verify} --controller <ip> --group <id> [--limit ip1,ip2] [--sudo-mode sudo|sudo-i]"
exit 1
fi
@ -66,7 +79,7 @@ if [[ "$ACTION" != "verify" && ( -z "$CONTROLLER_IP" || -z "$VTAP_GROUP_ID" ) ]]
fi
####################################
# 核心函数
# 核心函数（重点修改：SUDO 处理 + 重启逻辑）
####################################
worker() {
@ -74,7 +87,7 @@ worker() {
local user="$2"
local pass="$3"
echo "🔧 [$ACTION] 处理主机 $ip ($user)"
echo "<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> [$ACTION] 处理主机 $ip ($user)"
if [[ "$ACTION" == "verify" ]]; then
verify_agent "$ip" "$user" "$pass" && {
@ -109,13 +122,20 @@ worker() {
return
fi
install_agent "$ip" "$user" "$pass" "$pkg_path" && update_config "$ip" "$user" "$pass" && {
echo "✅ $ip $ACTION 完成"
echo "$ip" >> "$SUCCESS_FILE"
} || {
# 安装 + 配置
if install_agent "$ip" "$user" "$pass" "$pkg_path" && update_config "$ip" "$user" "$pass"; then
# ✅ 配置完成后，再次重启服务，确保新配置生效
restart_agent_service "$ip" "$user" "$pass" && {
echo "✅ $ip $ACTION 完成"
echo "$ip" >> "$SUCCESS_FILE"
} || {
echo "❌ $ip 服务重启失败"
echo "$ip" >> "$FAILED_FILE"
}
else
echo "❌ $ip 安装或配置失败"
echo "$ip" >> "$FAILED_FILE"
}
fi
echo "-------------------------------------------"
}
@ -164,75 +184,155 @@ choose_agent_package() {
if [[ ${#files[@]} -gt 0 ]]; then
latest=$(printf "%s\n" "${files[@]}" | sort -V | tail -1)
echo "🎯 选择安装包: $latest" >&2
echo "<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 选择安装包: $latest" >&2
echo "$latest"
else
echo "UNSUPPORTED"
fi
}
# ✅ 修改 install_agent：支持 sudo 和 sudo-i
install_agent() {
local ip="$1" user="$2" pass="$3" pkg_path="$4"
local remote_pkg="/tmp/agent.${pkg_path##*.}"
sshpass -p "$pass" scp $SSH_OPTS "$pkg_path" "$user@$ip:$remote_pkg"
sshpass -p "$pass" scp $SSH_OPTS "$pkg_path" "$user@$ip:$remote_pkg" || {
echo "❌ $ip 上传安装包失败"
return 1
}
# 构建 SUDO 前缀
local SUDO_CMD=""
case "$SUDO_MODE" in
sudo)
SUDO_CMD="sudo"
;;
sudo-i)
SUDO_CMD="sudo -i"
;;
*)
SUDO_CMD="sudo"
;;
esac
sshpass -p "$pass" ssh $SSH_OPTS "$user@$ip" bash <<EOF
set -e
if command -v sudo >/dev/null; then SUDO="sudo"; else SUDO=""; fi
# 使用指定模式
SUDO_MODE_CMD='$SUDO_CMD'
echo "<22><><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 使用权限模式: \$SUDO_MODE_CMD"
if [[ "$remote_pkg" == *.rpm ]]; then
\$SUDO rpm -Uvh --replacepkgs "$remote_pkg"
\$SUDO_MODE_CMD rpm -Uvh --replacepkgs "$remote_pkg"
elif [[ "$remote_pkg" == *.deb ]]; then
\$SUDO dpkg -i "$remote_pkg" || \$SUDO apt-get install -f -y
\$SUDO_MODE_CMD dpkg -i "$remote_pkg" || \$SUDO_MODE_CMD apt-get install -f -y
else
echo "❌ 不支持的安装包格式"
exit 1
fi
# 服务管理（注意：sudo -i 下 systemctl 可能仍可用）
if command -v systemctl &>/dev/null; then
\$SUDO systemctl enable $SERVICE_NAME
\$SUDO systemctl restart $SERVICE_NAME
\$SUDO_MODE_CMD systemctl enable $SERVICE_NAME
\$SUDO_MODE_CMD systemctl restart $SERVICE_NAME
elif command -v service &>/dev/null; then
\$SUDO service $SERVICE_NAME restart
\$SUDO chkconfig $SERVICE_NAME on
\$SUDO_MODE_CMD service $SERVICE_NAME restart
\$SUDO_MODE_CMD chkconfig $SERVICE_NAME on
elif command -v initctl &>/dev/null; then
\$SUDO initctl restart $SERVICE_NAME || \$SUDO initctl start $SERVICE_NAME
\$SUDO_MODE_CMD initctl restart $SERVICE_NAME || \$SUDO_MODE_CMD initctl start $SERVICE_NAME
else
echo "❌ 无法识别服务管理方式"
exit 1
fi
EOF
}
# ✅ 修改 update_config：确保配置写入 /etc/
update_config() {
local ip="$1" user="$2" pass="$3"
local SUDO_CMD=""
case "$SUDO_MODE" in
sudo)
SUDO_CMD="sudo"
;;
sudo-i)
SUDO_CMD="sudo -i"
;;
*)
SUDO_CMD="sudo"
;;
esac
sshpass -p "$pass" ssh $SSH_OPTS "$user@$ip" bash <<EOF
set -e
if command -v sudo >/dev/null; then SUDO="sudo"; else SUDO=""; fi
SUDO_MODE_CMD='$SUDO_CMD'
CONFIG_FILE="/etc/deepflow-agent.yaml"
\$SUDO mkdir -p \$(dirname \$CONFIG_FILE)
cat <<CFG | \$SUDO tee "\$CONFIG_FILE" >/dev/null
# 使用 sudo -i 创建目录（更可靠）
\$SUDO_MODE_CMD mkdir -p \$(dirname \$CONFIG_FILE)
# 写入配置（使用 tee 避免重定向权限问题）
cat <<'CFG' | \$SUDO_MODE_CMD tee "\$CONFIG_FILE" >/dev/null
controller-ips:
- $CONTROLLER_IP
vtap-group-id: "$VTAP_GROUP_ID"
CFG
\$SUDO chmod 644 "\$CONFIG_FILE"
\$SUDO chown root:root "\$CONFIG_FILE"
\$SUDO_MODE_CMD chmod 644 "\$CONFIG_FILE"
\$SUDO_MODE_CMD chown root:root "\$CONFIG_FILE"
EOF
}
# ✅ 新增函数：服务重启 + 状态检查
restart_agent_service() {
local ip="$1" user="$2" pass="$3"
local SUDO_CMD=""
case "$SUDO_MODE" in
sudo)
SUDO_CMD="sudo"
;;
sudo-i)
SUDO_CMD="sudo -i"
;;
*)
SUDO_CMD="sudo"
;;
esac
sshpass -p "$pass" ssh $SSH_OPTS "$user@$ip" bash <<EOF
set -e
SUDO_MODE_CMD='$SUDO_CMD'
# 重启服务
echo "<22><><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 正在重启 deepflow-agent.service 以应用配置..."
\$SUDO_MODE_CMD systemctl restart $SERVICE_NAME
sleep 3
# 检查服务状态
if ! \$SUDO_MODE_CMD systemctl is-active $SERVICE_NAME > /dev/null 2>&1; then
echo "❌ deepflow-agent.service 重启后未运行"
exit 1
fi
echo "✅ deepflow-agent.service 已成功重启"
EOF
}
verify_agent() {
local ip="$1" user="$2" pass="$3"
echo "🔍 $ip 状态检查："
echo "<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> $ip 状态检查："
sshpass -p "$pass" ssh $SSH_OPTS "$user@$ip" "
systemctl is-active $SERVICE_NAME 2>/dev/null || \
service $SERVICE_NAME status || \
initctl status $SERVICE_NAME
sudo systemctl is-active $SERVICE_NAME 2>/dev/null || \
sudo service $SERVICE_NAME status || \
sudo initctl status $SERVICE_NAME || \
echo '⚠️ 服务状态未知'
"
}
####################################
# 并发控制主逻辑
# 并发控制主逻辑（不变）
####################################
sem(){
@ -242,6 +342,8 @@ sem(){
}
while read -r ip user pass; do
[[ -z "$ip" || "$ip" =~ ^# ]] && continue
if [[ -n "$LIMIT" ]]; then
IFS=',' read -ra LIMIT_IPS <<< "$LIMIT"
skip=true
@ -260,7 +362,7 @@ wait
TOTAL_SUCCESS=$(wc -l < "$SUCCESS_FILE")
TOTAL_FAIL=$(wc -l < "$FAILED_FILE")
echo "🎯 全部任务执行完成: 成功 $TOTAL_SUCCESS 台，失败 $TOTAL_FAIL 台"
echo "<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 全部任务执行完成: 成功 $TOTAL_SUCCESS 台，失败 $TOTAL_FAIL 台"
if [[ -s "$FAILED_FILE" ]]; then
echo "❗ 失败主机列表已保存: $FAILED_FILE"
fi