#!/usr/bin/env bash # 重建并上线 governance api + web(本机 docker build 需要宿主代理 + --network host) set -u cd "$(dirname "${BASH_SOURCE[0]}")/.." || exit 1 [ -z "$(git status --porcelain --untracked-files=normal)" ] || { printf "%s\n" "REFUSED_DIRTY_SOURCE"; exit 1; } TS=$(date +%Y%m%d-%H%M%S) PROXY_ARGS=(--build-arg HTTP_PROXY=http://127.0.0.1:7890 --build-arg HTTPS_PROXY=http://127.0.0.1:7890 --build-arg NO_PROXY=localhost,127.0.0.1 --build-arg GOPROXY=https://goproxy.cn,direct) echo "=== 1. 打回滚标签 ($TS)" docker tag knowledge-governance-p1-api:latest "knowledge-governance-p1-api:rollback-$TS" || exit 1 docker tag knowledge-governance-p1-web:latest "knowledge-governance-p1-web:rollback-$TS" || exit 1 echo "=== 1a. 上线前校验 .env 的键值形状(一行一值)" # 2026-10-14 实测事故:`.env` 最后一行没有结尾换行、又被追加新键 ⇒ 两行粘成一行 # ⇒ VERSION_HISTORY_GITEA_BASE_URL 变成 "http://host:3000VERSION_HISTORY_LINK_ORIGIN=…" # ⇒ 配置校验按设计拒绝启动 ⇒ api/worker 崩溃重启(平台整体不可用)。 # 这里在**上线前**用形状判定拦住:值里再出现 `KEY=` 或键名不合法,直接拒绝部署。 if [ -f .env ]; then python3 - <<'ENVCHECK' || exit 1 import pathlib import re import sys problems = [] for number, raw in enumerate(pathlib.Path(".env").read_text(encoding="utf-8").splitlines(), 1): line = raw.strip() if not line or line.startswith("#") or "=" not in line: continue key, value = line.split("=", 1) if not re.fullmatch(r"[A-Za-z_][A-Za-z0-9_]*", key.strip()): problems.append((number, line, "键名不合法")) elif re.search(r"[A-Z][A-Z0-9_]{3,}=", value): # 不能用 \b:粘行处是 "…:3000VERSION_HISTORY_…",数字紧邻字母,没有词边界。 problems.append((number, line, "值里又出现 KEY=(两行粘在一起)")) if problems: print("ENV_VALUE_LOOKS_MERGED(.env 有坏值,拒绝上线)") for number, line, why in problems: print(f" line {number}: {why}: {line[:120]}") sys.exit(1) print(" .env 键值形状 OK") ENVCHECK fi echo "=== 1b. 构建身份(构建期注入:页面「当前运行」只认产物里的完整 SHA 精确匹配)" # 唯一来源是 scripts/version.py --identity(与 Makefile、CI 同一份规则):不注入 ⇒ 页面只能显示"未知"。 IDENTITY_ARGS=() while IFS= read -r line; do IDENTITY_ARGS+=("$line"); done < <(python3 scripts/version.py --identity 2>/dev/null | python3 -c ' import json, sys try: d = json.load(sys.stdin) except Exception: sys.exit(0) release = d.get("release_version") or "" count = d.get("commits_since_release") print("--build-arg=VERSION=" + (release or "dev")) print("--build-arg=RELEASE_VERSION=" + release) print("--build-arg=BUILD_KIND=" + (d.get("build_kind") or "development")) print("--build-arg=COMMITS_SINCE_RELEASE=" + ("" if count is None else str(count))) print("--build-arg=COMMIT_SHA=" + (d.get("commit_sha") or "unknown")) print("--build-arg=BUILD_TIME=" + (d.get("built_at") or "unknown")) ') if [ "${#IDENTITY_ARGS[@]}" -lt 6 ]; then echo "IDENTITY_UNAVAILABLE" echo " 取不到构建身份(scripts/version.py --identity 失败)⇒ 拒绝继续:" echo " 否则构建出来的产物无法被「版本记录」页识别成「当前运行」,只会显示未知。" exit 1 fi printf ' %s\n' "${IDENTITY_ARGS[@]}" echo "=== 1c. 保留上一版构建的静态资源(CDN/浏览器可能还持有旧 index.html)" # 只保留**当前正在跑的那一版**:旧 index.html 的旧哈希引用仍能取到文件, # 页面不会因为某个 JS/CSS 404 而空白或"裸奔"。目录只读挂载进 web 容器。 LEGACY_DIR="deploy/legacy-assets" PREV_WEB_IMAGE=$(docker inspect -f '{{.Image}}' knowledge-governance-p1-web-1 2>/dev/null || true) mkdir -p "$LEGACY_DIR/assets" && chmod 777 "$LEGACY_DIR" "$LEGACY_DIR/assets" if [ -n "$PREV_WEB_IMAGE" ]; then docker run --rm --user 0 --entrypoint sh -v "$PWD/$LEGACY_DIR:/out" "$PREV_WEB_IMAGE" \ -c 'cp -a /usr/share/nginx/html/assets/. /out/assets/ 2>/dev/null || true' >/dev/null 2>&1 || true echo " LEGACY_ASSETS=$(ls -1 "$LEGACY_DIR/assets" 2>/dev/null | tr '\n' ' ')" fi echo "=== 2. 构建 api" docker build --network host "${PROXY_ARGS[@]}" "${IDENTITY_ARGS[@]}" -f Dockerfile -t knowledge-governance-p1-api:latest . || { echo "API_BUILD_FAILED"; exit 1; } echo "=== 3. 构建 web" docker build --network host "${PROXY_ARGS[@]}" -f Dockerfile.web -t knowledge-governance-p1-web:latest . || { echo "WEB_BUILD_FAILED"; exit 1; } echo "=== 4. 上线 api + web(不动依赖)" # COMPOSE_FILES 让调用方叠加发布接线 / 门禁 override,默认只有基础 compose.yaml。 # 例:COMPOSE_FILES="-f compose.yaml -f compose.release-sim.yaml" bash scripts/deploy-api-web.sh # # 注意(2026-10-10 实测踩到):重建 api 时若不叠加 compose.release-gates.yaml,两道生产门禁会被 # **静默关掉**(PRODUCTION_RELEASE_REQUESTS_ENABLED/BACKUP_RESTORE_VERIFIED=false),页面只在点击 # 「授权发布」时回一句 "production release requests are disabled" —— 很容易误判成功能坏了。 # 因此下面做前置保护:正在运行的 api 若已开闸、而本次没叠加门禁文件,直接拒绝(确要关闸用 ALLOW_GATE_OFF=1)。 COMPOSE_FILES_EFFECTIVE="${COMPOSE_FILES:--f compose.yaml}" GATES_PREV=$(docker exec knowledge-governance-p1-api-1 sh -lc 'printf %s "${PRODUCTION_RELEASE_REQUESTS_ENABLED:-}"' 2>/dev/null || true) if [ "${GATES_PREV}" = "true" ] && [[ "${COMPOSE_FILES_EFFECTIVE}" != *compose.release-gates.yaml* ]] && [ "${ALLOW_GATE_OFF:-0}" != "1" ]; then echo "REFUSED_WOULD_CLOSE_PRODUCTION_GATES" echo " 当前 api 已开闸(PRODUCTION_RELEASE_REQUESTS_ENABLED=true),而本次 COMPOSE_FILES 不含 compose.release-gates.yaml," echo " 继续会用基础 compose 重建 api → 生产授权接口被静默关闭。" echo " 保留门禁:COMPOSE_FILES=\"-f compose.yaml -f compose.release-sim.yaml -f compose.release-gates.yaml\" bash scripts/deploy-api-web.sh" echo " 确实要关闸:ALLOW_GATE_OFF=1 bash scripts/deploy-api-web.sh" exit 1 fi docker compose ${COMPOSE_FILES_EFFECTIVE} up -d --no-build --no-deps api web || { echo "UP_FAILED"; exit 1; } echo "=== 5. 状态" docker ps --filter name=knowledge-governance-p1 --format '{{.Names}} {{.Status}} {{.Image}}' echo "api image=$(docker inspect -f '{{.Image}}' knowledge-governance-p1-api-1)" echo "web image=$(docker inspect -f '{{.Image}}' knowledge-governance-p1-web-1)" echo "=== 6. 健康" echo "health=$(curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:8088/health)" echo "=== 7. 生产端点环境变量(只报名字与非空)" docker exec knowledge-governance-p1-api-1 sh -lc 'for v in PRODUCTION_WEKNORA_BASE_URL PRODUCTION_WEKNORA_API_KEY_FILE PRODUCTION_WEKNORA_ALLOW_DOCKER_HTTP; do eval "val=\$$v"; if [ -n "$val" ]; then echo "$v="; else echo "$v="; fi; done' echo "=== 7b. 生产门禁状态(重建后必须与预期一致)" docker exec knowledge-governance-p1-api-1 sh -lc 'for v in PRODUCTION_RELEASE_REQUESTS_ENABLED PRODUCTION_BACKUP_RESTORE_VERIFIED; do eval "val=\$$v"; echo "$v=${val:-}"; done' echo "=== 8. 新包" docker exec knowledge-governance-p1-web-1 sh -lc 'grep -o "index-[A-Za-z0-9_-]*\.js" /usr/share/nginx/html/index.html | head -1' echo "=== 9. 构建身份是否真的进了产物(页面据此判「当前运行」)" INJECTED=$(printf '%s\n' "${IDENTITY_ARGS[@]}" | sed -n 's/^--build-arg=COMMIT_SHA=//p') if docker exec knowledge-governance-p1-api-1 sh -lc "strings /app/api | grep -q '$INJECTED'"; then echo "INJECTED_COMMIT_SHA=$INJECTED(已进 api 产物)" else echo "INJECTION_MISSING(产物里没有 $INJECTED —— 页面会显示当前运行未知)" fi echo "=== 10. 版本记录同步读回(当前 SHA + 注释快照 + 独立变更说明)" # Startup queues the read-only history job. A healthy API alone is not proof. VERIFIED=0 for ((attempt=0;attempt<90;attempt++)); do if docker exec knowledge-governance-p1-api-1 /app/api --version-record-status; then VERIFIED=1; break; fi sleep 2 done [ "$VERIFIED" = "1" ] || { printf '%s\n' 'VERSION_RECORD_NOT_VERIFIED'; exit 1; } echo "DONE"