fix: persist update logs and status

monitor/http_app.py

@@ -16,7 +16,7 @@ from .nacos import (
 )
 from .runtime_config import get_go_config_payload, save_go_config_payload, validate_go_config_payload
 from .service_release import build_update_targets_payload
-from .service_update_jobs import get_update_job_payload, start_update_job
+from .service_update_jobs import get_update_job_log_text, get_update_job_payload, start_update_job
 from .service_ops import build_restart_targets_payload, rolling_restart_payload
 
 
@@ -189,6 +189,30 @@ class MonitorHandler(BaseHTTPRequestHandler):
                 send_body=send_body,
             )
 
+        # 返回单个服务更新任务的完整文本日志。
+        if parsed.path == "/api/ops/update-log":
+            operation_id = self.query_value(query, "id")
+            if not operation_id:
+                return self.send_error_payload(
+                    HTTPStatus.BAD_REQUEST,
+                    "id is required",
+                    send_body=send_body,
+                )
+
+            try:
+                content = get_update_job_log_text(operation_id)
+            except KeyError:
+                return self.send_error_payload(HTTPStatus.NOT_FOUND, "update log not found", send_body=send_body)
+            except Exception as exc:  # noqa: BLE001
+                return self.send_error_payload(HTTPStatus.BAD_GATEWAY, str(exc), send_body=send_body)
+
+            return self.send_payload(
+                HTTPStatus.OK,
+                content.encode("utf-8"),
+                "text/plain; charset=utf-8",
+                send_body=send_body,
+            )
+
         # favicon 直接返回空。
         if parsed.path == "/favicon.ico":
             return self.send_payload(HTTPStatus.NO_CONTENT, b"", "image/x-icon", send_body=send_body)

monitor/service_release.py

@@ -27,6 +27,7 @@ from .config import (
     HARBOR_REGISTRY,
     HARBOR_USERNAME,
     JAVA_REPO_ROOT,
+    ROOT,
     ROLLING_RESTART_STABILIZE_SECONDS,
     ROLLING_RESTART_TIMEOUT_SECONDS,
     SERVICE_BUILD_TIMEOUT_SECONDS,
@@ -75,6 +76,8 @@ SERVICE_IMAGE_REPOSITORY_SUFFIXES = {
 
 # 后台任务的日志回调统一走这个签名。
 ProgressLogger = Callable[[str, str, str], None]
+# 私有 Maven 安装缓存统一放到 run 目录，避免混进 Git 工作区。
+PRIVATE_MAVEN_CACHE_DIR = ROOT / "run" / "maven_private_cache"
 
 
 def emit_progress(progress: ProgressLogger | None, stage: str, message: str, *, level: str = "info") -> None:
@@ -272,6 +275,77 @@ def fetch_repo_refs(repo_root: Path, label: str) -> None:
     )
 
 
+def local_maven_repo_root() -> Path:
+    # 兼容外部显式指定 MAVEN_REPO_LOCAL，默认仍沿用 ~/.m2/repository。
+    override = str(os.environ.get("MAVEN_REPO_LOCAL") or "").strip()
+    if override:
+        return Path(override).expanduser()
+    return Path.home() / ".m2" / "repository"
+
+
+def private_maven_tree_id(repo_root: Path, commit: str) -> str:
+    # 用 git tree id 表示 maven_private 内容，只有私有依赖目录变化时才会变化。
+    try:
+        return local_command_output(
+            ["git", "-C", str(repo_root), "rev-parse", "--verify", f"{commit}:maven_private"],
+            cwd=repo_root,
+            label="resolve private maven tree",
+        )
+    except Exception:  # noqa: BLE001
+        return ""
+
+
+def private_maven_cache_stamp_path(repo_root: Path) -> Path:
+    # 每个仓保留一份最近已安装的私有依赖 tree id。
+    return PRIVATE_MAVEN_CACHE_DIR / f"{repo_root.name}.tree"
+
+
+def private_maven_artifacts_present() -> bool:
+    # 本地 Maven 仓至少得有一份私有制品，缓存命中才有意义。
+    artifacts_root = local_maven_repo_root() / "com" / "red" / "circle"
+    if not artifacts_root.exists():
+        return False
+    for path in artifacts_root.rglob("*.pom"):
+        if path.is_file():
+            return True
+    return False
+
+
+def ensure_private_maven_installed(
+    repo_root: Path,
+    worktree_root: Path,
+    commit: str,
+    *,
+    progress: ProgressLogger | None = None,
+) -> None:
+    # 私有依赖没变化时直接复用本机 ~/.m2 缓存，避免每次更新都重复 install-file。
+    tree_id = private_maven_tree_id(repo_root, commit)
+    stamp_path = private_maven_cache_stamp_path(repo_root)
+    cached_tree_id = stamp_path.read_text(encoding="utf-8").strip() if stamp_path.exists() else ""
+
+    if tree_id and cached_tree_id == tree_id and private_maven_artifacts_present():
+        emit_progress(progress, "build.java.maven", f"私有 Maven 依赖未变化，跳过安装：{tree_id[:12]}")
+        return
+
+    emit_progress(progress, "build.java.maven", "开始安装私有 Maven 依赖")
+    run_logged_command(
+        [str(worktree_root / "ci" / "install-private-maven.sh")],
+        cwd=worktree_root,
+        label="java install private maven",
+        timeout_seconds=min(SERVICE_BUILD_TIMEOUT_SECONDS, 1200),
+        progress_stage="build.java.maven.install",
+        progress=progress,
+    )
+
+    if tree_id:
+        PRIVATE_MAVEN_CACHE_DIR.mkdir(parents=True, exist_ok=True)
+        stamp_path.write_text(f"{tree_id}\n", encoding="utf-8")
+        emit_progress(progress, "build.java.maven", f"私有 Maven 依赖安装完成：{tree_id[:12]}")
+        return
+
+    emit_progress(progress, "build.java.maven", "私有 Maven 依赖安装完成")
+
+
 def resolve_repo_commit(repo_root: Path, git_ref: str, label: str) -> str:
     # 优先把分支解析到 origin/<branch>，避免本地旧分支 HEAD 干扰构建。
     candidates = [
@@ -440,13 +514,10 @@ def build_java_images(
 
     try:
         if not skip_maven_build:
-            emit_progress(progress, "build.java.maven", "开始安装私有 Maven 依赖")
-            run_logged_command(
-                [str(worktree_root / "ci" / "install-private-maven.sh")],
-                cwd=worktree_root,
-                label="java install private maven",
-                timeout_seconds=min(SERVICE_BUILD_TIMEOUT_SECONDS, 1200),
-                progress_stage="build.java.maven.install",
+            ensure_private_maven_installed(
+                repo_root,
+                worktree_root,
+                commit,
                 progress=progress,
             )
             emit_progress(progress, "build.java.maven", f"开始 Maven 打包：{maven_module_list(service_names)}")
@@ -479,7 +550,8 @@ def build_java_images(
                 label=f"java build {service_name}",
                 env={
                     "DOCKER_PLATFORM": BUILD_DOCKER_PLATFORM,
-                    "SKIP_MAVEN_BUILD": "1" if skip_maven_build else "0",
+                    # 外层已经统一决定过是否 Maven 打包；镜像阶段一律复用现成产物。
+                    "SKIP_MAVEN_BUILD": "1",
                 },
                 timeout_seconds=SERVICE_BUILD_TIMEOUT_SECONDS,
                 progress_stage=f"build.java.{service_name}.build",

monitor/service_update_jobs.py

@@ -2,13 +2,18 @@ from __future__ import annotations
 
 # 返回状态给 HTTP 层时直接用 deepcopy 最稳。
 import copy
+# 状态持久化到 JSON 文件需要标准库 json。
+import json
+# 日志和状态文件落到 run 目录。
+from pathlib import Path
 # 后台更新任务需要用线程跑，避免 HTTP 请求一直阻塞。
 import threading
 # operationId 用 uuid 生成最直接。
+import re
 import uuid
 from typing import Any
 
-from .config import utc_now
+from .config import ROOT, utc_now
 from .service_release import deploy_updated_services_payload
 
 
@@ -16,6 +21,8 @@ from .service_release import deploy_updated_services_payload
 MAX_UPDATE_OPERATIONS = 20
 # 单次更新最多保留最近这些日志行，超出后丢弃最旧的。
 MAX_UPDATE_LOG_LINES = 1200
+# 任务状态和完整日志统一落到 run 目录。
+UPDATE_JOB_DIR = ROOT / "run" / "service_update_jobs"
 # 统一保护任务状态的锁。
 UPDATE_JOB_LOCK = threading.Lock()
 # 当前进程内的更新任务表。
@@ -30,6 +37,126 @@ def normalize_services(service_names: list[str]) -> list[str]:
     return list(dict.fromkeys(normalized))
 
 
+def update_job_state_path(operation_id: str) -> Path:
+    # 每个任务一份状态 JSON。
+    return UPDATE_JOB_DIR / f"{operation_id}.json"
+
+
+def update_job_log_path(operation_id: str) -> Path:
+    # 每个任务一份完整文本日志。
+    return UPDATE_JOB_DIR / f"{operation_id}.log"
+
+
+def ensure_update_job_dir() -> None:
+    # 持久化目录不存在时先创建。
+    UPDATE_JOB_DIR.mkdir(parents=True, exist_ok=True)
+
+
+def valid_operation_id(operation_id: str) -> bool:
+    # 只接受简单安全字符，避免路径拼接被滥用。
+    return re.fullmatch(r"[0-9A-Za-z_-]{6,64}", str(operation_id or "").strip()) is not None
+
+
+def append_log_file(operation_id: str, line: str) -> None:
+    # 完整日志单独写入文本文件，供页面下载。
+    ensure_update_job_dir()
+    with update_job_log_path(operation_id).open("a", encoding="utf-8") as handle:
+        handle.write(line)
+
+
+def operation_payload(payload: dict[str, Any]) -> dict[str, Any]:
+    # 对外返回和落盘都统一走这个结构，避免前后不一致。
+    return {
+        "operationId": payload["operationId"],
+        "kind": payload["kind"],
+        "status": payload["status"],
+        "stage": payload["stage"],
+        "message": payload["message"],
+        "error": payload["error"],
+        "startedAt": payload["startedAt"],
+        "updatedAt": payload["updatedAt"],
+        "finishedAt": payload["finishedAt"],
+        "logDroppedCount": payload["logDroppedCount"],
+        "logs": copy.deepcopy(payload.get("logs") or []),
+        "result": copy.deepcopy(payload.get("result")),
+        "request": copy.deepcopy(payload.get("request") or {}),
+        "logUrl": f"/api/ops/update-log?id={payload['operationId']}",
+    }
+
+
+def persist_job_unlocked(payload: dict[str, Any]) -> None:
+    # 每次状态变化都落盘，服务重启后还能恢复最近任务。
+    ensure_update_job_dir()
+    state_path = update_job_state_path(payload["operationId"])
+    state_path.write_text(
+        json.dumps({"ok": True, "operation": operation_payload(payload)}, ensure_ascii=False, indent=2) + "\n",
+        encoding="utf-8",
+    )
+
+
+def recover_lost_running_job_unlocked(payload: dict[str, Any]) -> None:
+    # 进程重启后，原后台线程已经不存在；把悬挂中的 running/queued 任务收口成失败态最直接。
+    if payload.get("status") not in {"queued", "running"}:
+        return
+
+    now = utc_now()
+    message = "monitor process restarted before update status could be finalized"
+    payload["status"] = "failed"
+    payload["stage"] = "failed"
+    payload["message"] = "service update interrupted"
+    payload["error"] = message
+    payload["updatedAt"] = now
+    payload["finishedAt"] = now
+    payload["logs"].append(
+        {
+            "time": now,
+            "stage": "failed",
+            "level": "error",
+            "message": message,
+        }
+    )
+    if len(payload["logs"]) > MAX_UPDATE_LOG_LINES:
+        overflow = len(payload["logs"]) - MAX_UPDATE_LOG_LINES
+        payload["logs"] = payload["logs"][overflow:]
+        payload["logDroppedCount"] = int(payload.get("logDroppedCount") or 0) + overflow
+
+    append_log_file(payload["operationId"], f"[{now}] [failed] [error] {message}\n")
+    persist_job_unlocked(payload)
+
+
+def load_job_from_disk(operation_id: str) -> dict[str, Any] | None:
+    # 任务不在内存时，从持久化状态恢复。
+    if not valid_operation_id(operation_id):
+        return None
+
+    state_path = update_job_state_path(operation_id)
+    if not state_path.exists():
+        return None
+
+    payload = json.loads(state_path.read_text(encoding="utf-8"))
+    operation = dict(payload.get("operation") or {})
+    if not operation:
+        return None
+
+    restored = {
+        "operationId": str(operation.get("operationId") or operation_id).strip(),
+        "kind": str(operation.get("kind") or "service-update").strip() or "service-update",
+        "status": str(operation.get("status") or "").strip(),
+        "stage": str(operation.get("stage") or "").strip(),
+        "message": str(operation.get("message") or "").strip(),
+        "error": str(operation.get("error") or "").strip(),
+        "startedAt": str(operation.get("startedAt") or "").strip(),
+        "updatedAt": str(operation.get("updatedAt") or "").strip(),
+        "finishedAt": str(operation.get("finishedAt") or "").strip(),
+        "logDroppedCount": int(operation.get("logDroppedCount") or 0),
+        "logs": list(operation.get("logs") or []),
+        "result": operation.get("result"),
+        "request": dict(operation.get("request") or {}),
+    }
+    recover_lost_running_job_unlocked(restored)
+    return restored
+
+
 def create_job_payload(
     service_names: list[str],
     *,
@@ -64,8 +191,11 @@ def create_job_payload(
         },
     }
     with UPDATE_JOB_LOCK:
+        ensure_update_job_dir()
         UPDATE_JOBS[operation_id] = payload
         UPDATE_JOB_ORDER.append(operation_id)
+        update_job_log_path(operation_id).write_text("", encoding="utf-8")
+        persist_job_unlocked(payload)
         trim_jobs_unlocked()
     return snapshot_job(operation_id)
 
@@ -102,6 +232,11 @@ def append_job_log(operation_id: str, stage: str, message: str, *, level: str =
             overflow = len(payload["logs"]) - MAX_UPDATE_LOG_LINES
             payload["logs"] = payload["logs"][overflow:]
             payload["logDroppedCount"] = int(payload.get("logDroppedCount") or 0) + overflow
+        append_log_file(
+            operation_id,
+            f"[{payload['updatedAt']}] [{str(stage or '').strip() or payload['stage']}] [{str(level or 'info').strip() or 'info'}] {text}\n",
+        )
+        persist_job_unlocked(payload)
 
 
 def patch_job(operation_id: str, **fields: Any) -> None:
@@ -112,6 +247,7 @@ def patch_job(operation_id: str, **fields: Any) -> None:
             return
         payload.update(fields)
         payload["updatedAt"] = utc_now()
+        persist_job_unlocked(payload)
 
 
 def finish_job(operation_id: str, *, result: dict[str, Any], status: str, error: str = "") -> None:
@@ -131,6 +267,7 @@ def finish_job(operation_id: str, *, result: dict[str, Any], status: str, error:
         payload["finishedAt"] = now
         payload["updatedAt"] = now
         payload["result"] = result
+        persist_job_unlocked(payload)
 
 
 def snapshot_job(operation_id: str) -> dict[str, Any]:
@@ -140,24 +277,7 @@ def snapshot_job(operation_id: str) -> dict[str, Any]:
         if payload is None:
             raise KeyError(operation_id)
 
-        return {
-            "ok": True,
-            "operation": {
-                "operationId": payload["operationId"],
-                "kind": payload["kind"],
-                "status": payload["status"],
-                "stage": payload["stage"],
-                "message": payload["message"],
-                "error": payload["error"],
-                "startedAt": payload["startedAt"],
-                "updatedAt": payload["updatedAt"],
-                "finishedAt": payload["finishedAt"],
-                "logDroppedCount": payload["logDroppedCount"],
-                "logs": copy.deepcopy(payload.get("logs") or []),
-                "result": copy.deepcopy(payload.get("result")),
-                "request": copy.deepcopy(payload.get("request") or {}),
-            },
-        }
+        return {"ok": True, "operation": operation_payload(payload)}
 
 
 def job_progress_logger(operation_id: str):
@@ -246,4 +366,27 @@ def start_update_job(
 
 def get_update_job_payload(operation_id: str) -> dict[str, Any]:
     # 供 HTTP 轮询接口直接读取任务状态。
-    return snapshot_job(operation_id)
+    try:
+        return snapshot_job(operation_id)
+    except KeyError:
+        restored = load_job_from_disk(operation_id)
+        if restored is None:
+            raise
+
+        with UPDATE_JOB_LOCK:
+            UPDATE_JOBS[operation_id] = restored
+            if operation_id not in UPDATE_JOB_ORDER:
+                UPDATE_JOB_ORDER.append(operation_id)
+            trim_jobs_unlocked()
+        return snapshot_job(operation_id)
+
+
+def get_update_job_log_text(operation_id: str) -> str:
+    # 页面下载完整日志时直接返回文本文件内容。
+    if not valid_operation_id(operation_id):
+        raise KeyError(operation_id)
+
+    log_path = update_job_log_path(operation_id)
+    if not log_path.exists():
+        raise KeyError(operation_id)
+    return log_path.read_text(encoding="utf-8", errors="replace")

static/app.js

@@ -333,6 +333,13 @@ createApp({
     updateTargetLabel(item) {
       return `${item.service} · ${item.kind} · ${item.hosts.map((host) => host.host).join(" / ")}`;
     },
+    updateLogDownloadUrl() {
+      if (!this.updateOps.operationId) {
+        return "";
+      }
+      const query = new URLSearchParams({ id: this.updateOps.operationId });
+      return `/api/ops/update-log?${query.toString()}`;
+    },
     updateStatusBadgeClass() {
       if (this.updateOps.error || this.updateOps.status === "failed") {
         return "bad";
@@ -468,11 +475,30 @@ createApp({
           }
         }
       } catch (error) {
+        const errorMessage = error instanceof Error ? error.message : String(error);
+        const notFound = errorMessage.includes("update operation not found") || errorMessage.includes("HTTP 404");
+
+        if (silent && !notFound) {
+          return;
+        }
+
         this.stopUpdatePolling();
         this.updateOps.running = false;
-        if (!silent) {
-          this.updateOps.error = error instanceof Error ? error.message : String(error);
+
+        if (notFound) {
+          this.updateOps.status = "failed";
+          this.updateOps.error = "更新任务状态不存在，可能因为服务重启或旧任务已清理";
+          this.updateOps.message = "";
+          this.updateOps.operationId = "";
+          this.persistUpdateOperationId();
+          return;
         }
+
+        if (this.updateOps.status === "queued" || this.updateOps.status === "running") {
+          this.updateOps.status = "failed";
+          this.updateOps.message = "更新状态查询失败";
+        }
+        this.updateOps.error = errorMessage;
       }
     },
     applyRefreshTimer() {

static/index.html

@@ -501,16 +501,28 @@
                 <div class="panel-head update-log-head">
                     <div>
                         <h3>Update Log</h3>
-                        <p>按阶段展示构建、推 Harbor、同步 compose、预拉镜像和滚动重建的实时输出</p>
+                        <p>按阶段展示构建、推 Harbor、同步 compose、预拉镜像和滚动重建的实时输出，可下载完整日志</p>
+                    </div>
+                    <div class="action-group">
+                        <button class="refresh ghost" @click="fetchUpdateOperationStatus(updateOps.operationId)" :disabled="!updateOps.operationId || updateOps.running">
+                            刷新日志
+                        </button>
+                        <a
+                            v-if="updateOps.operationId"
+                            class="refresh ghost"
+                            :href="updateLogDownloadUrl()"
+                            target="_blank"
+                            rel="noopener"
+                            download
+                        >
+                            下载完整日志
+                        </a>
                     </div>
-                    <button class="refresh ghost" @click="fetchUpdateOperationStatus(updateOps.operationId)" :disabled="!updateOps.operationId || updateOps.running">
-                        刷新日志
-                    </button>
                 </div>
                 <pre id="service-update-log-output" class="update-log-output mono">{{ updateLogText || '等待更新日志...' }}</pre>
             </section>
 
-            <div v-if="updateOps.result && updateOps.result.build && updateOps.result.build.length > 0" class="table-wrap">
+            <div v-if="updateOps.result && updateOps.result.builds && updateOps.result.builds.length > 0" class="table-wrap">
                 <table class="table compact">
                     <thead>
                         <tr>
@@ -522,7 +534,7 @@
                         </tr>
                     </thead>
                     <tbody>
-                        <tr v-for="build in updateOps.result.build" :key="build.kind + '-' + build.commit">
+                        <tr v-for="build in updateOps.result.builds" :key="build.kind + '-' + build.commit">
                             <td>{{ build.kind }}</td>
                             <td class="mono">{{ build.gitRef }}</td>
                             <td class="mono">{{ build.commit }}</td>