自定义制作docker容器自动自愈容器镜像

包括：

• 完整的 autoheal.sh（支持每分钟检查一次、连续 5 次 unhealthy 才重启）
• Dockerfile
• docker-compose.yml
• 详细文档，包含参数说明、用法

---

1️⃣ autoheal.sh

#!/usr/bin/env sh
set -e
set -o pipefail

DOCKER_SOCK=${DOCKER_SOCK:-/var/run/docker.sock}
CURL_TIMEOUT=${CURL_TIMEOUT:-30}
WEBHOOK_URL=${WEBHOOK_URL:-""}

HTTP_ENDPOINT="http://localhost"
UNIX_SOCK="--unix-socket ${DOCKER_SOCK}"

AUTOHEAL_CONTAINER_LABEL=${AUTOHEAL_CONTAINER_LABEL:-autoheal}
AUTOHEAL_START_PERIOD=${AUTOHEAL_START_PERIOD:-0}
AUTOHEAL_INTERVAL=${AUTOHEAL_INTERVAL:-60}     # 默认每分钟检查一次
AUTOHEAL_DEFAULT_STOP_TIMEOUT=${AUTOHEAL_DEFAULT_STOP_TIMEOUT:-10}
AUTOHEAL_MAX_FAILS=${AUTOHEAL_MAX_FAILS:-5}    # 连续5次才重启

STATE_FILE="/tmp/autoheal_counts"
touch "$STATE_FILE"

docker_curl() {
  curl --max-time "${CURL_TIMEOUT}" --no-buffer -s \
  ${UNIX_SOCK} \
  "$@"
}

get_container_info() {
  local label_filter
  if [ "$AUTOHEAL_CONTAINER_LABEL" = "all" ]; then
    label_filter=""
  else
    label_filter=",\"label\":\[\"${AUTOHEAL_CONTAINER_LABEL}=true\"\]"
  fi
  url="${HTTP_ENDPOINT}/containers/json?filters=\{\"health\":\[\"unhealthy\"\]${label_filter}\}"
  docker_curl "$url"
}

restart_container() {
  local container_id="$1"
  local timeout="$2"
  docker_curl -f -X POST "${HTTP_ENDPOINT}/containers/${container_id}/restart?t=${timeout}"
}

get_count() {
  grep "^$1 " "$STATE_FILE" | awk '{print $2}'
}

set_count() {
  grep -v "^$1 " "$STATE_FILE" > "${STATE_FILE}.tmp" 2>/dev/null || true
  echo "$1 $2" >> "${STATE_FILE}.tmp"
  mv "${STATE_FILE}.tmp" "$STATE_FILE"
}

trap 'kill $$; exit 143' SIGTERM

if [ "$1" = "autoheal" ] && [ -e "$DOCKER_SOCK" ]; then
  [ "$AUTOHEAL_START_PERIOD" -gt 0 ] && sleep "$AUTOHEAL_START_PERIOD"

  while true; do
    STOP_TIMEOUT=".Labels[\"autoheal.stop.timeout\"] // $AUTOHEAL_DEFAULT_STOP_TIMEOUT"
    get_container_info | \
      jq -r "foreach .[] as \$C([];[];\$C | .Id, .Names[0], .State, ${STOP_TIMEOUT})" | \
      while read -r CONTAINER_ID && read -r CONTAINER_NAME && read -r CONTAINER_STATE && read -r TIMEOUT; do
        CONTAINER_SHORT_ID=${CONTAINER_ID:0:12}
        DATE=$(date +%F" "%T)
        COUNT=$(get_count "$CONTAINER_ID")
        COUNT=${COUNT:-0}

        if [ "$CONTAINER_NAME" = "null" ]; then
          echo "$DATE Container ${CONTAINER_SHORT_ID} name null, skip"
        elif [ "$CONTAINER_STATE" = "restarting" ]; then
          echo "$DATE Container $CONTAINER_NAME ($CONTAINER_SHORT_ID) restarting, skip"
        else
          COUNT=$((COUNT + 1))
          echo "$DATE Container $CONTAINER_NAME ($CONTAINER_SHORT_ID) unhealthy count=$COUNT"
          if [ $COUNT -ge $AUTOHEAL_MAX_FAILS ]; then
            echo "$DATE Container $CONTAINER_NAME ($CONTAINER_SHORT_ID) reached $AUTOHEAL_MAX_FAILS, restarting"
            if restart_container "$CONTAINER_ID" "$TIMEOUT"; then
              echo "$DATE Restarted container $CONTAINER_SHORT_ID ok"
            else
              echo "$DATE Restarted container $CONTAINER_SHORT_ID failed" >&2
            fi
            COUNT=0
          fi
          set_count "$CONTAINER_ID" "$COUNT"
        fi
      done
    sleep "$AUTOHEAL_INTERVAL"
  done
else
  exec "$@"
fi

---

2️⃣ Dockerfile

FROM alpine:3.19
RUN apk add --no-cache bash curl jq

COPY autoheal.sh /usr/local/bin/autoheal.sh
RUN chmod +x /usr/local/bin/autoheal.sh

# 自身容器健康检查
HEALTHCHECK --interval=5s --timeout=3s --retries=3 CMD pgrep -f autoheal || exit 1

ENTRYPOINT ["/usr/local/bin/autoheal.sh"]
CMD ["autoheal"]

构建：

docker build -t my-autoheal .

---

3️⃣ docker-compose.yml

version: "3.8"
services:
  autoheal:
    image: my-autoheal:latest
    container_name: autoheal
    restart: unless-stopped
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
    environment:
      # === 参数示例 ===
      - DOCKER_SOCK=/var/run/docker.sock
      - AUTOHEAL_CONTAINER_LABEL=all
      - AUTOHEAL_START_PERIOD=0
      - AUTOHEAL_INTERVAL=60
      - AUTOHEAL_MAX_FAILS=5
      - AUTOHEAL_DEFAULT_STOP_TIMEOUT=10
      # - WEBHOOK_URL=https://your-webhook

启动：

docker-compose up -d

---

4️⃣ 参数说明

环境变量	说明	默认值
DOCKER_SOCK	Docker Daemon 的 socket 或 TCP 地址	/var/run/docker.sock
AUTOHEAL_CONTAINER_LABEL	要监控的容器标签：autoheal（只监控打了 autoheal=true 标签的容器），all（所有容器）	autoheal
AUTOHEAL_START_PERIOD	容器启动后延迟多少秒再开始监控	0
AUTOHEAL_INTERVAL	检查间隔秒数	60
AUTOHEAL_MAX_FAILS	连续多少次 unhealthy 才重启	5
AUTOHEAL_DEFAULT_STOP_TIMEOUT	停止容器的超时时间（秒）	10
WEBHOOK_URL	重启成功/失败时发送通知的 Webhook URL（可选）	空

如果用 TCP 远程 Docker API，还可以设置证书路径（自己扩展 CA_CERT、CLIENT_CERT、CLIENT_KEY 等）。

---

5️⃣ 使用示例

你的业务容器：

services:
  myapp:
    image: myapp:latest
    labels:
      - autoheal=true  # 如果 AUTOHEAL_CONTAINER_LABEL=autoheal 时

再用上面的 autoheal 服务即可实现：

• 每分钟检查一次容器健康
• 连续 5 次 unhealthy 才重启
• 重启成功/失败可通过 Webhook 通知