fix(deploy): 修复组件版本解析和安装升级验收流程

2026-10-18 00:18:00 +08:00
parent 895d92317e
commit b99eb849b4
30 changed files with 1579 additions and 270 deletions
@@ -0,0 +1,83 @@
+#!/bin/bash
+# Shared installation/update checks. Never print compose config or credentials.
+init_deployment_docker_config() {
+    local file="$1" configured fallback
+    configured="$(read_env_value "$file" DOCKER_CONFIG_FILE)"
+    if [ -n "$configured" ] && [ -f "$configured" ]; then return 0; fi
+    if [ -n "$configured" ] && [ "$configured" != /root/.docker/config.json ]; then
+        echo '❌ DOCKER_CONFIG_FILE 不是可读取的配置文件；未启动服务' >&2
+        return 1
+    fi
+    if [ -f "${DOCKER_CONFIG:-$HOME/.docker}/config.json" ]; then
+        write_env_value "$file" DOCKER_CONFIG_FILE "${DOCKER_CONFIG:-$HOME/.docker}/config.json"
+        return 0
+    fi
+    # Anonymous installations have no ~/.docker/config.json. A missing short
+    # bind source would otherwise be created by Docker as a DIRECTORY.
+    fallback="$PWD/data/docker-auth/config.json"
+    mkdir -p "$(dirname "$fallback")"
+    chmod 700 "$(dirname "$fallback")"
+    if [ ! -e "$fallback" ]; then (umask 077; printf '{}\n' > "$fallback"); fi
+    [ -f "$fallback" ] || { echo '❌ 安装目录的镜像认证配置不是文件' >&2; return 1; }
+    write_env_value "$file" DOCKER_CONFIG_FILE "$fallback"
+}
+
+verify_deployment_images() {
+    local compose="$1" configuration
+    shift
+    configuration="$(sudo $compose config --format json)" || return 1
+    printf '%s' "$configuration" | python3 ./scripts/verify-deployment-images.py --sudo --services "$@"
+}
+
+check_deployment_storage() {
+    local compose="$1" configuration
+    shift
+    configuration="$(sudo $compose config --format json)" || return 1
+    printf '%s' "$configuration" | python3 ./scripts/check-deployment-storage.py --sudo --services "$@"
+}
+
+verify_pulled_release() {
+    local compose="$1" configuration
+    shift
+    configuration="$(sudo $compose config --format json)" || return 1
+    printf '%s' "$configuration" | python3 ./scripts/verify-pulled-release.py --sudo --services "$@"
+}
+
+wait_for_deployment_services() {
+    local compose="$1" timeout="$2" deadline service container_id state running health all_ready passes=0
+    shift 2
+    local services=("$@")
+    [[ "$timeout" =~ ^[1-9][0-9]*$ ]] || { echo '❌ 健康检查等待秒数必须为正整数' >&2; return 1; }
+    if [ "${#services[@]}" -eq 0 ]; then
+        while IFS= read -r service; do [ -z "$service" ] || services+=("$service"); done < <(sudo $compose config --services)
+    fi
+    [ "${#services[@]}" -gt 0 ] || return 1
+    deadline=$((SECONDS + timeout))
+    echo "⏳ 等待 ${#services[@]} 个服务恢复可用（最长 ${timeout} 秒）..."
+    while (( SECONDS < deadline )); do
+        all_ready=true
+        for service in "${services[@]}"; do
+            container_id="$(sudo $compose ps -q "$service" 2>/dev/null || true)"
+            if [ -z "$container_id" ]; then all_ready=false; continue; fi
+            state="$(sudo docker inspect --format '{{.State.Running}}|{{if .State.Health}}{{.State.Health.Status}}{{else}}none{{end}}' "$container_id" 2>/dev/null || true)"
+            running="${state%%|*}"; health="${state#*|}"
+            if [ "$running" != true ] || { [ "$health" != none ] && [ "$health" != healthy ]; }; then
+                all_ready=false
+            elif [ "$service" = watchtower ] && [ "$health" = none ]; then
+                sudo docker exec "$container_id" /watchtower --health-check >/dev/null 2>&1 || all_ready=false
+            fi
+        done
+        if [ "$all_ready" = true ]; then
+            passes=$((passes + 1))
+            if [ "$passes" -ge 2 ]; then echo '✅ 所选服务已恢复可用'; return 0; fi
+            sleep 2
+        else
+            passes=0
+            sleep 5
+        fi
+    done
+    echo '❌ 服务未在规定时间内恢复可用，部署未完成' >&2
+    sudo $compose ps "${services[@]}" || true
+    echo "请运行：$compose logs --tail=100 ${services[*]}" >&2
+    return 1
+}