|
-
+
+
+
+
|
{c.jobType[job.type]}
@@ -1505,6 +1552,12 @@ export function JobDetailPage({
>;
}) {
const zh = c.nav.overview === "总览";
+ const [jobIdCopied, setJobIdCopied] = useState(false);
+ const handleCopyResourceId = async () => {
+ if (!(await copyText(job.id))) return;
+ setJobIdCopied(true);
+ window.setTimeout(() => setJobIdCopied(false), 1600);
+ };
const [activeTab, setActiveTab] = useState<"workers" | "logs" | "metrics">(
"workers",
);
@@ -1533,9 +1586,28 @@ export function JobDetailPage({
logs: string;
}>
>([]);
+ const [backendLogs, setBackendLogs] = useState<
+ Array<{
+ timestamp: string;
+ line: string;
+ labels?: Record;
+ fields?: Record;
+ }>
+ >([]);
const [logsLoading, setLogsLoading] = useState(false);
const [logsError, setLogsError] = useState(null);
- const [workerRoleFilter, setWorkerRoleFilter] = useState("All");
+ const [logsHasMore, setLogsHasMore] = useState(false);
+ const [logsNextCursor, setLogsNextCursor] = useState("");
+ const [logsLoadingMore, setLogsLoadingMore] = useState(false);
+ // 游标历史栈，用于上一页/下一页翻页。首页为空字符串，第一页查完后 push(nextCursor)。
+ const [logsCursorHistory, setLogsCursorHistory] = useState([""]);
+ const [logsPageIndex, setLogsPageIndex] = useState(0);
+ // 角色配置区块的角色选择（默认第一个角色）
+ const [workerRoleFilter, setWorkerRoleFilter] = useState(
+ job.resources.length > 0 ? job.resources[0].role : "All",
+ );
+ // Worker 列表的角色筛选（独立于角色配置区块，默认 All）
+ const [workerListRoleFilter, setWorkerListRoleFilter] = useState("All");
const [workerPage, setWorkerPage] = useState(1);
const [workerSort, setWorkerSort] = useState<{
key:
@@ -1545,6 +1617,8 @@ export function JobDetailPage({
| "node"
| "kind"
| "ip"
+ | "domainIP"
+ | "gpu"
| "createdAt"
| "phase";
direction: SortDirection;
@@ -1558,6 +1632,9 @@ export function JobDetailPage({
const [logWorkerFilter, setLogWorkerFilter] = useState("All");
const [logQuery, setLogQuery] = useState("");
const [logRange, setLogRange] = useState("1h");
+ const [logCustomRange, setLogCustomRange] = useState(false);
+ const [logCustomFrom, setLogCustomFrom] = useState("");
+ const [logCustomTo, setLogCustomTo] = useState("");
const [logStreamEnabled, setLogStreamEnabled] = useState(false);
// Aggregate Node CR pullProgress for the top StatusBadge hover. Uses Node CR
@@ -1625,10 +1702,18 @@ export function JobDetailPage({
});
const workerTaskNamesKey = workerTaskNames.join(",");
+ const handleWorkerRefresh = async () => {
+ if (workerRefreshing) return;
+ setWorkerRefreshing(true);
+ await refreshTasks();
+ setWorkerRefreshKey((key) => key + 1);
+ };
+
useEffect(() => {
if (!workerTaskNamesKey) {
setPods([]);
setDomainIPMap({});
+ setWorkerRefreshing(false);
return;
}
let cancelled = false;
@@ -1737,69 +1822,180 @@ export function JobDetailPage({
[pendingPodNamesKey],
);
- const fetchLogs = async (isInitial = true) => {
+ const getLogTimeRange = (): { from: string; to: string } => {
+ if (logCustomRange) {
+ if (logCustomFrom && logCustomTo) {
+ // Convert local datetime-local input (no timezone) to UTC ISO string
+ const fromDate = new Date(logCustomFrom);
+ const toDate = new Date(logCustomTo);
+ return { from: fromDate.toISOString(), to: toDate.toISOString() };
+ }
+ // Fallback to 1h if custom range is incomplete
+ }
+ const to = new Date();
+ const from = new Date();
+ switch (logRange) {
+ case "15m":
+ from.setMinutes(from.getMinutes() - 15);
+ break;
+ case "1h":
+ from.setHours(from.getHours() - 1);
+ break;
+ case "6h":
+ from.setHours(from.getHours() - 6);
+ break;
+ case "24h":
+ from.setHours(from.getHours() - 24);
+ break;
+ case "7d":
+ from.setDate(from.getDate() - 7);
+ break;
+ case "30d":
+ from.setDate(from.getDate() - 30);
+ break;
+ default:
+ from.setHours(from.getHours() - 1);
+ }
+ // Convert to UTC ISO string (new Date() is already local time)
+ return { from: from.toISOString(), to: to.toISOString() };
+ };
+
+ const fetchLogs = async (isInitial = true, cursor = "") => {
if (activeTab !== "logs") return;
- if (!isInitial && !logStreamEnabled) return;
if (isInitial) setLogsLoading(true);
+ if (cursor) setLogsLoadingMore(true);
setLogsError(null);
try {
+ const params = new URLSearchParams();
+ const { from, to } = getLogTimeRange();
+ params.set("from", from);
+ params.set("to", to);
+
+ // Always send the first task as the base filter (backend requires it)
+ let taskName = "";
+ if (logRoleFilter !== "All") {
+ const resource = job.resources.find((r) => r.role === logRoleFilter);
+ taskName = resource
+ ? taskResourceName(job.name, resource.role)
+ : logRoleFilter;
+ } else if (logWorkerFilter !== "All") {
+ const matchedPod = pods.find((p) => p.podName === logWorkerFilter);
+ if (matchedPod && matchedPod.taskName) {
+ taskName = matchedPod.taskName;
+ }
+ }
+
+ if (!taskName && job.resources.length > 0) {
+ // Default to the first resource's task if nothing selected
+ taskName = taskResourceName(job.name, job.resources[0].role);
+ }
+
+ if (taskName) {
+ params.set("task", taskName);
+ }
+
+ if (logWorkerFilter !== "All") {
+ params.set("pod", logWorkerFilter);
+ }
+
+ if (logQuery.trim()) {
+ params.set("query", logQuery.trim());
+ }
+
+ if (cursor) {
+ params.set("cursor", cursor);
+ }
+
const resp = await fetch(
- `/api/v1/rlinf.io/v1alpha1/jobs/${encodeURIComponent(job.name)}/logs`,
+ `/api/v1/rlinf.io/v1alpha1/jobs/${encodeURIComponent(job.name)}/logs?${params.toString()}`,
);
if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
const data = await resp.json();
- setPodLogs(Array.isArray(data.pods) ? data.pods : []);
+ if (data.source === "backend" && Array.isArray(data.entries)) {
+ // 分页模式：替换当前页数据，而不是追加
+ setBackendLogs(data.entries);
+ setPodLogs([]);
+ setLogsHasMore(Boolean(data.hasMore));
+ setLogsNextCursor(data.nextCursor || "");
+ } else {
+ setPodLogs(Array.isArray(data.pods) ? data.pods : []);
+ setBackendLogs([]);
+ setLogsHasMore(false);
+ setLogsNextCursor("");
+ }
} catch (e) {
- setPodLogs([]);
+ if (!cursor) {
+ setPodLogs([]);
+ setBackendLogs([]);
+ }
setLogsError(e instanceof Error ? e.message : String(e));
} finally {
setLogsLoading(false);
+ setLogsLoadingMore(false);
}
};
- useAutoRefresh(fetchLogs, 5000, [activeTab, job.name, logStreamEnabled]);
+ // 查询条件变化时重置到第一页
+ const resetLogsPagination = () => {
+ setLogsCursorHistory([""]);
+ setLogsPageIndex(0);
+ };
- const fallbackWorkers: WorkerItem[] =
- job.taskStatuses.length > 0
- ? job.taskStatuses.map((ts, i) => {
- const childTaskName = ts.name.toLowerCase().replace(/\s+/g, "-");
- const jobChildName = `${job.name}-${childTaskName}`
- .toLowerCase()
- .replace(/\s+/g, "-");
- return {
- id: `${job.id}-${i}`,
- name: ts.name,
- jobId: job.id,
- role: ts.name,
- node:
- taskNodes[jobChildName] ??
- taskNodes[ts.name] ??
- ts.observedNodes?.join(", ") ??
- "—",
- cluster: taskClusters[jobChildName] ?? taskClusters[ts.name] ?? "—",
- phase: (ts.phase || "Pending") as Phase,
- cpu: job.resources.find((item) => item.role === ts.name)?.cpu ?? "",
- memory:
- job.resources.find((item) => item.role === ts.name)?.memory ?? "",
- gpu: job.resources.find((item) => item.role === ts.name)?.gpu,
- logs: ts.message
- ? [ts.message]
- : [
- `${ts.name}: worker state synced`,
- `${ts.name}: waiting for runtime heartbeat`,
- ],
- statusMessage: ts.message || undefined,
- pullProgress:
- pullProgressMap[jobChildName] ??
- pullProgressMap[ts.name.toLowerCase()] ??
- [],
- events:
- taskEventsMap[jobChildName] ??
- taskEventsMap[ts.name.toLowerCase()] ??
- [],
- };
- })
- : [];
+ const goToNextLogPage = () => {
+ if (!logsHasMore || !logsNextCursor || logsLoadingMore) return;
+ const nextHistory = [...logsCursorHistory];
+ nextHistory[logsPageIndex + 1] = logsNextCursor;
+ setLogsCursorHistory(nextHistory);
+ setLogsPageIndex(logsPageIndex + 1);
+ fetchLogs(false, logsNextCursor);
+ };
+
+ const goToPrevLogPage = () => {
+ if (logsPageIndex === 0 || logsLoadingMore) return;
+ const prevIndex = logsPageIndex - 1;
+ const prevCursor = logsCursorHistory[prevIndex] || "";
+ setLogsPageIndex(prevIndex);
+ fetchLogs(false, prevCursor);
+ };
+
+ // 查询条件变化时重置分页到第一页
+ useEffect(() => {
+ resetLogsPagination();
+ }, [
+ logRange,
+ logCustomRange,
+ logCustomFrom,
+ logCustomTo,
+ logRoleFilter,
+ logWorkerFilter,
+ logQuery,
+ ]);
+
+ useAutoRefresh(fetchLogs, 5000, [
+ activeTab,
+ job.name,
+ logStreamEnabled,
+ logRange,
+ logCustomRange,
+ logCustomFrom,
+ logCustomTo,
+ logRoleFilter,
+ logWorkerFilter,
+ logQuery,
+ ]);
+
+ const schedulingSummary = (
+ message: string | undefined,
+ events: NodeEventEntry[],
+ ) =>
+ message === "FailedScheduling" ||
+ events.some((event) => event.reason === "FailedScheduling")
+ ? zh
+ ? "没有合适的节点可调度，资源可能被占用"
+ : "No suitable node is available; resources may be occupied"
+ : message;
+
+ const fallbackWorkers: WorkerItem[] = [];
const resourceForTask = (taskName: string) =>
job.resources.find(
(resource) => taskResourceName(job.name, resource.role) === taskName,
@@ -1815,7 +2011,9 @@ export function JobDetailPage({
// hover surfaces live image pull progress for this worker.
const nodePullProgress =
phase === "Pending" && pod.node
- ? (nodePullProgressMap[pod.node] ?? [])
+ ? (nodePullProgressMap[pod.node] ?? []).filter(
+ (progress) => progress.image === resource?.image,
+ )
: [];
// 同样从 Node.status.events 取节点 warning 事件；当 pod.node
// 缺失时回退到 Task.status.events，让 Pending worker 行 tooltip
@@ -1845,7 +2043,7 @@ export function JobDetailPage({
`${role}: worker state synced`,
`${role}: waiting for runtime heartbeat`,
],
- statusMessage: pod.message || undefined,
+ statusMessage: schedulingSummary(pod.message, workerEvents),
pullProgress: nodePullProgress,
events: workerEvents,
};
@@ -1854,10 +2052,7 @@ export function JobDetailPage({
const runningWorkerCount = jobWorkers.filter(
(worker) => worker.phase === "Running",
).length;
- const displayPhase = effectiveJobPhase(
- job,
- jobWorkers.map((worker) => worker.phase),
- );
+ const displayPhase = effectiveJobPhase(job);
const workerPodsByTask = new Map();
for (const worker of jobWorkers) {
workerPodsByTask.set(
@@ -1869,9 +2064,12 @@ export function JobDetailPage({
),
);
}
- const workerRoles = [...new Set(jobWorkers.map((worker) => worker.role))];
+ const workerRoles = [
+ ...new Set(job.resources.map((resource) => resource.role)),
+ ];
const filteredWorkers = jobWorkers.filter(
- (worker) => workerRoleFilter === "All" || worker.role === workerRoleFilter,
+ (worker) =>
+ workerListRoleFilter === "All" || worker.role === workerListRoleFilter,
);
const toggleWorkerSort = (key: typeof workerSort.key) => {
setWorkerSort((current) => ({
@@ -1899,6 +2097,14 @@ export function JobDetailPage({
return getNodeKindLabel(worker);
case "ip":
return pod?.ip ?? "";
+ case "domainIP":
+ return worker.id
+ ? (domainIPMap[
+ `${worker.id.split("/")[0]}/${worker.id.split("/")[1]}/${worker.id.split("/")[2]}`
+ ] ?? "")
+ : "";
+ case "gpu":
+ return worker.gpu ?? "";
case "createdAt":
return formatWorkerCreatedAt(job.startedAt, index);
case "phase":
@@ -1953,33 +2159,74 @@ export function JobDetailPage({
workerTableDrag.current.active = false;
setWorkerTableDragging(false);
};
- const logEntries = podLogs.flatMap((pod) => {
- const role =
- resourceForTask(pod.taskName)?.role ??
- job.resources.find((resource) => resource.role === pod.taskName)?.role ??
- pod.taskName;
- return pod.logs
- .split("\n")
- .filter(Boolean)
- .map((message, index) => ({
- id: `${pod.podName}-${index}-${message}`,
- worker: pod.podName,
- role,
- phase: pod.phase,
- node: pod.node,
- message,
- }));
- });
- const logRoles = [...new Set(logEntries.map((entry) => entry.role))];
- const logWorkers = [
- ...new Set(
- logEntries
- .filter(
- (entry) => logRoleFilter === "All" || entry.role === logRoleFilter,
- )
- .map((entry) => entry.worker),
- ),
- ];
+ const logEntries =
+ backendLogs.length > 0
+ ? backendLogs.map((entry, index) => {
+ const taskName = entry.labels?.task || entry.fields?.task || "";
+ const podName = entry.labels?.pod || entry.fields?.pod || "";
+ const node = entry.labels?.node || entry.fields?.node || "";
+ const role =
+ resourceForTask(taskName)?.role ??
+ job.resources.find((resource) => resource.role === taskName)
+ ?.role ??
+ taskName;
+ return {
+ id: `${podName}-${entry.timestamp}-${index}`,
+ worker: podName,
+ role,
+ phase: "Running", // Backend logs don't have phase, assume running
+ node,
+ message: entry.line,
+ timestamp: entry.timestamp,
+ };
+ })
+ : podLogs.flatMap((pod) => {
+ const role =
+ resourceForTask(pod.taskName)?.role ??
+ job.resources.find((resource) => resource.role === pod.taskName)
+ ?.role ??
+ pod.taskName;
+ return pod.logs
+ .split("\n")
+ .filter(Boolean)
+ .map((message, index) => ({
+ id: `${pod.podName}-${index}-${message}`,
+ worker: pod.podName,
+ role,
+ phase: pod.phase,
+ node: pod.node,
+ message,
+ timestamp: undefined,
+ }));
+ });
+ // Derive available roles and workers from Job metadata (resources and pods)
+ // so the dropdowns remain populated and selectable even when log query returns 0 entries.
+ const logRoles = useMemo(() => {
+ const rolesFromResources = job.resources.map((r) => r.role);
+ const rolesFromPods = pods.map((p) => {
+ const resource = resourceForTask(p.taskName);
+ return resource?.role ?? p.taskName;
+ });
+ return [...new Set([...rolesFromResources, ...rolesFromPods])].filter(
+ Boolean,
+ );
+ }, [job.resources, pods]);
+
+ const logWorkers = useMemo(() => {
+ return [
+ ...new Set(
+ pods
+ .filter((p) => {
+ if (logRoleFilter === "All") return true;
+ const resource = resourceForTask(p.taskName);
+ const role = resource?.role ?? p.taskName;
+ return role === logRoleFilter;
+ })
+ .map((p) => p.podName),
+ ),
+ ].filter(Boolean);
+ }, [pods, logRoleFilter]);
+
const filteredLogEntries = logEntries.filter(
(entry) =>
(logRoleFilter === "All" || entry.role === logRoleFilter) &&
@@ -1989,7 +2236,7 @@ export function JobDetailPage({
.includes(logQuery.toLowerCase()),
);
const tabs: Array<{ id: typeof activeTab; label: string }> = [
- { id: "workers", label: zh ? "Worker" : "Workers" },
+ { id: "workers", label: zh ? "详情" : "Details" },
{ id: "logs", label: c.common.logs },
{ id: "metrics", label: zh ? "监控" : "Metrics" },
];
@@ -2002,374 +2249,29 @@ export function JobDetailPage({
: undefined;
return (
-
-
- {tabs.map((tab) => (
-
- ))}
-
- {activeTab === "workers" && (
-
-
-
-
- {zh ? "运行实例" : "Runtime instances"}
-
- {zh ? "Worker 列表" : "Worker list"}
-
-
-
- {filteredWorkers.length} {zh ? "个 Worker" : "workers"}
-
-
-
-
- {
- setWorkerRoleFilter(role);
- setWorkerPage(1);
- }}
- />
-
-
-
-
- {(
- [
- ["name", zh ? "实例名称" : "Worker name"],
- ["role", zh ? "角色" : "Role"],
- ["cluster", zh ? "集群" : "Cluster"],
- ["node", zh ? "节点" : "Node"],
- ["kind", zh ? "节点类型" : "Node type"],
- ["ip", zh ? "实例 IP" : "Worker IP"],
- ["createdAt", zh ? "创建时间" : "Created"],
- ["phase", zh ? "状态" : "Status"],
- ] as const
- ).map(([key, label]) => (
- |
- toggleWorkerSort(key)}
- />
- |
- ))}
- |
-
-
-
- {visibleWorkers.map(({ worker, index }) => (
-
- ))}
-
-
-
-
- {zh
- ? `第 ${workerPage} / ${workerPageCount} 页`
- : `Page ${workerPage} of ${workerPageCount}`}
-
-
-
-
-
-
- )}
- {activeTab === "logs" && (
-
-
-
- {zh ? "任务日志" : "Job logs"}
- {zh ? "Worker 日志流" : "Worker log stream"}
-
-
- {logsLoading ? (
-
-
-
-
-
-
- {zh ? "正在连接 Worker 日志" : "Connecting to worker logs"}
-
-
- {zh
- ? "正在汇总各实例的最新输出…"
- : "Collecting the latest output from each instance…"}
-
-
-
-
- ) : logsError ? (
- {logsError}
- ) : podLogs.length === 0 ? (
- {zh ? "暂无日志" : "No logs available"}
- ) : (
- <>
-
-
-
-
-
-
-
-
-
- {filteredLogEntries.length > 0 ? (
- <>
-
- {zh ? "角色" : "Role"}
- Worker
- {zh ? "日志内容" : "Log message"}
-
- {filteredLogEntries.map((entry) => (
-
- {entry.role}
- {entry.worker}
- {entry.message}
-
- ))}
-
- ) : (
-
- {zh ? "未找到匹配的日志。" : "No matching logs found."}
-
- )}
-
-
- )}
-
- )}
- {activeTab === "metrics" && (
-
-
-
- {zh ? "任务监控" : "Job metrics"}
-
- {zh
- ? "Worker 资源与具身通道"
- : "Worker resources and live channel"}
-
-
-
-
-
- )}
-
{c.jobs.selected}
- {job.name}
-
- {c.jobType[job.type]} · {job.cluster}
-
+ {job.displayName}
+
+
+
+ {c.jobType[job.type]} · {job.cluster}
+
+
@@ -2474,114 +2376,608 @@ function JobPublicOverview({
-
-
-
-
-
-
-
-
-
- {zh ? "访问入口" : "Access"}
- Header Worker
-
- {tensorBoardProxy && (
-
+ {/* 第一块：角色配置信息（平铺展示，支持切换角色） */}
+ {
+ setWorkerRoleFilter(role);
+ setWorkerPage(1);
+ }}
+ />
+
+ {/* 第二块：公共配置 */}
+
+
+ {/* 第三块：Pod 实例列表 */}
+
+
+
+
+ {zh ? "Worker 列表" : "Runtime workers"}
+
+ {zh ? "Worker 维度查看实例信息" : "Worker list"}
+
+
+
+
+ {workerRoles.map((role) => (
+
+ ))}
+
+
+ {filteredWorkers.length} {zh ? "个 Pod" : "pods"}
+
+
+
+
+
-
-
- )}
-
-
- {job.headerRole || "—"}
- {job.headerWorker && job.headerWorker !== job.headerRole && (
- {job.headerWorker}
- )}
-
-
+
+
+
+ {(
+ [
+ ["name", zh ? "实例名称" : "Worker name"],
+ ["role", zh ? "角色" : "Role"],
+ ["cluster", zh ? "集群" : "Cluster"],
+ ["node", zh ? "节点" : "Node"],
+ ["kind", zh ? "节点类型" : "Node type"],
+ ["ip", zh ? "实例 IP" : "Worker IP"],
+ ["domainIP", zh ? "网络域 IP" : "Domain IP"],
+ ["gpu", zh ? "申请 GPU" : "GPU"],
+ ["createdAt", zh ? "创建时间" : "Created"],
+ ["phase", zh ? "状态" : "Status"],
+ ] as const
+ ).map(([key, label]) => (
+ |
+ toggleWorkerSort(key)}
+ />
+ |
+ ))}
+
+ {zh ? "操作" : "Actions"}
+ |
+
+
+
+ {visibleWorkers.length > 0 ? (
+ visibleWorkers.map(({ worker, index }) => (
+
+ ))
+ ) : (
+
+ |
+ {zh
+ ? "当前没有运行中的 Worker 实例"
+ : "No running worker instances"}
+ |
+
+ )}
+
+
+
+
+ {zh
+ ? `第 ${workerPage} / ${workerPageCount} 页`
+ : `Page ${workerPage} of ${workerPageCount}`}
+
+
+
+
+
+
-
-
-
- {zh ? "启动命令" : "Start command"}
-
-
+ {activeTab === "logs" && (
+
+
+
+ {zh ? "任务日志" : "Job logs"}
+ {zh ? "Worker 日志流" : "Worker log stream"}
+
-
-
- {zh ? "公共配置" : "Shared settings"}
-
-
- {baseConfigRows.map((row) => (
-
- {row.label}
- {row.value}
+ {logsError ? (
+ {logsError}
+ ) : (
+ <>
+
+
+
+
+ {logCustomRange && (
+
+ setLogCustomFrom(e.target.value)}
+ />
+
+ 至
+
+ setLogCustomTo(e.target.value)}
+ />
+
+ )}
+
+
+
+
+
+ {logsLoading ? (
+
+
+
+
+
+
+ {zh
+ ? "正在连接 Worker 日志"
+ : "Connecting to worker logs"}
+
+
+ {zh
+ ? "正在汇总各实例的最新输出…"
+ : "Collecting the latest output from each instance…"}
+
+
+
- ))}
+ ) : (
+
+ {filteredLogEntries.length > 0 ? (
+ <>
+
+ {zh ? "角色" : "Role"}
+ Worker
+ {zh ? "日志内容" : "Log message"}
+ {zh ? "时间" : "Time"}
+
+ {filteredLogEntries.map((entry) => (
+
+ {entry.role}
+
+ {entry.worker}
+
+ {entry.message}
+
+ {entry.timestamp
+ ? new Date(entry.timestamp).toLocaleString(
+ zh ? "zh-CN" : "en-US",
+ {
+ year: "numeric",
+ month: "2-digit",
+ day: "2-digit",
+ hour: "2-digit",
+ minute: "2-digit",
+ second: "2-digit",
+ hour12: false,
+ },
+ )
+ : ""}
+
+
+ ))}
+ {backendLogs.length > 0 && (
+
+
+
+ {zh
+ ? `第 ${logsPageIndex + 1} 页`
+ : `Page ${logsPageIndex + 1}`}
+
+
+
+ )}
+
+ ) : (
+
+ {zh ? "未找到匹配的日志。" : "No matching logs found."}
+
+ )}
+
+ )}
+
+ )}
+
+ )}
+
+ {activeTab === "metrics" && (
+
+
+
+
+ {zh ? "任务监控" : "Job metrics"}
+
+
+ {zh
+ ? "Worker 资源与具身通道"
+ : "Worker resources and live channel"}
+
-
-
- ({
- key: item.key,
- value: item.value,
- }))
- : []
- }
- empty={
- zh ? "未配置环境变量" : "No environment variables configured"
- }
+
+ )}
+
+
+ );
+}
+
+function JobPublicOverview({
+ job,
+ copy: c,
+ onBack,
+ runningWorkerCount,
+ totalWorkers,
+ displayPhase,
+ tensorBoardProxy,
+ onClone,
+ lifecycleActions,
+ jobPullProgress = [],
+ jobEvents = [],
+ jobFailedMessage,
+}: {
+ job: Job;
+ copy: CopyType;
+ onBack: () => void;
+ runningWorkerCount: number;
+ totalWorkers: number;
+ displayPhase: JobDisplayPhase;
+ tensorBoardProxy?: string;
+ onClone?: () => void;
+ lifecycleActions: JobLifecycleActions;
+ jobPullProgress?: PullProgressEntry[];
+ jobEvents?: NodeEventEntry[];
+ jobFailedMessage?: string;
+}) {
+ const zh = c.nav.overview === "总览";
+ const baseConfigRows = [
+ {
+ label: zh ? "Worker 数量" : "Worker count",
+ value: `${runningWorkerCount} / ${totalWorkers}`,
+ },
+ {
+ label: zh ? "创建时间" : "Created",
+ value: formatTaskTime(job.startedAt),
+ },
+ {
+ label: "Header Worker",
+ value: job.headerRole || "—",
+ },
+ {
+ label: zh ? "网络域" : "Network domain",
+ value: job.domain || (zh ? "未配置" : "Not configured"),
+ },
+ {
+ label: "TensorBoard",
+ value: job.tensorBoardDir || (zh ? "未配置" : "Not configured"),
+ },
+ {
+ label: zh ? "SSH 公钥" : "SSH Public Key",
+ value: job.sshPublicKey
+ ? `${job.sshPublicKey.slice(0, 32)}...`
+ : zh
+ ? "未配置"
+ : "Not configured",
+ fullValue: job.sshPublicKey,
+ },
+ ];
+ return (
+
+
+
+
+ {zh ? "公共配置" : "Shared configuration"}
+
+
+ {zh ? "任务维度查看公共配置" : "Job-level attributes configuration"}
+
+
+
+
+
+
+ {zh ? "启动命令" : "Start command"}
+
+
+
+
+
+ {zh ? "公共属性" : "Shared attributes"}
+
+
+ {baseConfigRows.map((row) => (
+
+ {row.label}
+ {row.value}
+
+ ))}
+
+
@@ -2742,7 +3138,6 @@ function RoleRuntimeConfig({
>;
}) {
const zh = c.nav.overview === "总览";
- const [expanded, setExpanded] = useState(false);
const resource = job.resources.find((item) => item.role === selectedRole);
const taskName = resource ? taskResourceName(job.name, resource.role) : "";
const taskStatus = job.taskStatuses.find(
@@ -2793,21 +3188,17 @@ function RoleRuntimeConfig({
.filter(Boolean)
.join(" / ") || (zh ? "未申请设备" : "No device requested")
: "";
- const nodeSelectors = resource?.nodeSelector
- ? resource.nodeSelector.split(",").map((selector) => {
- const [key, ...value] = selector.split("=");
- return { key: key.trim(), value: value.join("=").trim() };
- })
- : [];
return (
-
+
- {zh ? "Worker 角色" : "Worker roles"}
+
+ {zh ? "角色配置" : "Roles configuration"}
+
{zh
- ? "按角色查看实例与运行配置"
+ ? "角色维度查看运行配置"
: "Instances and configuration by role"}
@@ -2815,23 +3206,11 @@ function RoleRuntimeConfig({
className="role-runtime-tabs"
aria-label={zh ? "选择角色" : "Select role"}
>
-
{roles.map((role) => (
-
) : (
@@ -2888,7 +3250,7 @@ function RoleRuntimeConfig({
: `Showing workers from all ${roles.length} roles`}
)}
- {resource && expanded && (
+ {resource && (
@@ -2901,14 +3263,12 @@ function RoleRuntimeConfig({
/>
- {zh ? "节点选择" : "Node selectors"}
- {nodeSelectors.length ? (
+ {zh ? "节点选择" : "Node selection"}
+ {resourceNodes.length ? (
- {nodeSelectors.map((selector, index) => (
-
- {selector.key}
- =
- {selector.value || "—"}
+ {resourceNodes.map((node, index) => (
+
+ {node}
))}
@@ -3615,6 +4007,18 @@ function WorkerTableRow({
{pod?.ip || "—"}
|
+
+ {domainIP}
+ |
+
+
+ {worker.gpu && worker.gpu !== "0"
+ ? `${worker.gpu} GPU`
+ : zh
+ ? "未申请"
+ : "None"}
+
+ |
{createdAt}
|
@@ -3648,7 +4052,7 @@ function WorkerTableRow({
)}
-
+ |
-
-
-
|
- {expanded && (
-
-
-
-
-
-
- {zh ? "Worker 详情" : "Worker details"}
-
- {worker.name}
-
- {sshCommand && (
-
-
- {sshCommand}
-
-
- )}
-
-
-
- {zh ? "集群" : "Cluster"}
-
-
-
- {zh ? "角色" : "Role"}
- {worker.role}
-
-
- {zh ? "节点" : "Node"}
-
-
-
- {zh ? "申请 GPU" : "GPU request"}
-
- {worker.gpu && worker.gpu !== "0"
- ? worker.gpu
- : zh
- ? "未申请"
- : "Not requested"}
-
-
-
- {pods.length > 0 ? (
-
-
-
-
- | {zh ? "实例名称" : "Worker name"} |
- {zh ? "节点" : "Node"} |
- {zh ? "实例 IP" : "Worker IP"} |
- {zh ? "网络域" : "Domain"} |
- {zh ? "状态" : "Status"} |
-
-
-
- {pods.map((pod) => {
- const domainIP =
- domainIPMap[
- `${pod.namespace}/${pod.podNamespace}/${pod.podName}`
- ] ?? "";
- const podPhase = (pod.phase || "Pending") as Phase;
- // Mirror the worker-row tooltip aggregation so the pod
- // subtable surfaces the same failure reason / image-pull
- // progress / node events as the top-right status badge.
- const podPullProgress =
- podPhase === "Pending" && pod.node
- ? (nodePullProgressMap[pod.node] ?? [])
- : [];
- const podEvents =
- podPhase === "Pending"
- ? (podEventsMap[pod.name] ?? []).length > 0
- ? (podEventsMap[pod.name] ?? [])
- : pod.node &&
- (nodeEventsMap[pod.node] ?? []).length > 0
- ? (nodeEventsMap[pod.node] ?? [])
- : (taskEventsMap[
- pod.taskName?.toLowerCase() ?? ""
- ] ?? [])
- : [];
- return (
-
-
- {pod.podName}
- |
-
-
- |
- {pod.ip || "—"} |
-
- {pod.domain ? (
- <>
-
- {pod.domain}
- {domainIP && (
-
- {domainIP}
-
- )}
-
- ) : (
- "—"
- )}
- |
-
-
-
- {podPhase !== "Running" &&
- (podPhase === "Pending" ||
- podPhase === "Failed" ||
- podPullProgress.length > 0 ||
- podEvents.length > 0) && (
-
- )}
- |
-
- );
- })}
-
-
-
- ) : (
-
- {zh ? "暂无 Pod 详情，等待任务同步。" : "No pod details yet."}
-
- )}
- |
-
- )}
);
}
diff --git a/apps/rlark-ui/src/pages/Overview.tsx b/apps/rlark-ui/src/pages/Overview.tsx
index 7d32597..de5a661 100644
--- a/apps/rlark-ui/src/pages/Overview.tsx
+++ b/apps/rlark-ui/src/pages/Overview.tsx
@@ -42,6 +42,7 @@ export function Overview({
const [realClusters, setRealClusters] = useState([]);
const [realNodes, setRealNodes] = useState([]);
const [realJobs, setRealJobs] = useState([]);
+ const [refreshing, setRefreshing] = useState(false);
const isZh = c.nav.overview === "总览";
const { refresh } = useAutoRefresh(async () => {
@@ -62,6 +63,16 @@ export function Overview({
setRealJobs(jobItems.map(crdToJob));
}, 15000);
+ const handleRefresh = async () => {
+ if (refreshing) return;
+ setRefreshing(true);
+ try {
+ await refresh();
+ } finally {
+ setRefreshing(false);
+ }
+ };
+
const displayNodes = realNodes;
const embodiedClusters = realClusters.filter((x) => x.type === "Embodied");
const runningJobs = realJobs.filter((x) => x.phase === "Running").length;
@@ -306,8 +317,18 @@ export function Overview({
{c.overview.recent}
{c.common.production}
-
diff --git a/apps/rlark-ui/src/pages/SSHKeys.tsx b/apps/rlark-ui/src/pages/SSHKeys.tsx
index c247f61..99e0dc8 100644
--- a/apps/rlark-ui/src/pages/SSHKeys.tsx
+++ b/apps/rlark-ui/src/pages/SSHKeys.tsx
@@ -162,8 +162,21 @@ export function SSHKeysPage({ copy: c }: { copy: Copy }) {
className="secondary-button"
onClick={fetchKeys}
title={zh ? "刷新" : "Refresh"}
+ aria-label={zh ? "刷新" : "Refresh"}
+ aria-busy={loading}
+ disabled={loading}
>
-
+
+ {loading
+ ? zh
+ ? "刷新中..."
+ : "Refreshing..."
+ : zh
+ ? "刷新"
+ : "Refresh"}
-
diff --git a/apps/rlark-ui/src/pages/SystemConfig.tsx b/apps/rlark-ui/src/pages/SystemConfig.tsx
index 4e541f0..ce141a4 100644
--- a/apps/rlark-ui/src/pages/SystemConfig.tsx
+++ b/apps/rlark-ui/src/pages/SystemConfig.tsx
@@ -1,10 +1,31 @@
import { useEffect, useState } from "react";
-import { Settings, Save, Check, Copy as CopyIcon } from "lucide-react";
+import {
+ Settings,
+ Save,
+ Check,
+ Copy as CopyIcon,
+ RefreshCw,
+} from "lucide-react";
import type { Copy } from "../i18n";
+interface LogBackendConfig {
+ endpoint: string;
+ project: string;
+ logstore: string;
+ accessKeyId: string;
+ accessKeySecret: string;
+}
+
+interface LogConfig {
+ backend: string;
+ config: LogBackendConfig;
+}
+
interface SystemConfig {
sshJumpHost: string;
sshJumpPort: string;
+ log: LogConfig;
+ isAccessKeySecretSet: boolean; // 标记 accessKeySecret 是否已设置（用于区分掩码和用户输入）
}
export function SystemConfigPage({ copy: c }: { copy: Copy }) {
@@ -12,12 +33,25 @@ export function SystemConfigPage({ copy: c }: { copy: Copy }) {
const [config, setConfig] = useState ({
sshJumpHost: "",
sshJumpPort: "",
+ log: {
+ backend: "sls",
+ config: {
+ endpoint: "",
+ project: "",
+ logstore: "",
+ accessKeyId: "",
+ accessKeySecret: "",
+ },
+ },
+ isAccessKeySecretSet: false,
});
const [loading, setLoading] = useState(true);
const [saving, setSaving] = useState(false);
const [error, setError] = useState("");
const [saved, setSaved] = useState(false);
const [copied, setCopied] = useState(false);
+ // 记录日志配置是否被修改过，用于决定是否在保存时发送 log 字段
+ const [isLogConfigDirty, setIsLogConfigDirty] = useState(false);
useEffect(() => {
fetchConfig();
@@ -30,10 +64,25 @@ export function SystemConfigPage({ copy: c }: { copy: Copy }) {
const resp = await fetch("/api/v1/system-config");
if (!resp.ok) throw new Error(await resp.text());
const data = await resp.json();
+ const secret = data.log?.config?.accessKeySecret || "";
setConfig({
- sshJumpHost: data.sshJumpHost || "",
- sshJumpPort: data.sshJumpPort || "",
+ sshJumpHost: data.ssh?.jumpHost || data.sshJumpHost || "",
+ sshJumpPort: data.ssh?.jumpPort || data.sshJumpPort || "",
+ log: {
+ backend: data.log?.backend || "sls",
+ config: {
+ endpoint: data.log?.config?.endpoint || "",
+ project: data.log?.config?.project || "",
+ logstore: data.log?.config?.logstore || "",
+ accessKeyId: data.log?.config?.accessKeyId || "",
+ // 后端返回掩码时展示掩码；用于让用户知道已设置
+ accessKeySecret: secret,
+ },
+ },
+ isAccessKeySecretSet: Boolean(secret),
});
+ // 加载完成后，重置脏标记
+ setIsLogConfigDirty(false);
} catch (e) {
setError(String(e));
} finally {
@@ -46,10 +95,22 @@ export function SystemConfigPage({ copy: c }: { copy: Copy }) {
setError("");
setSaved(false);
try {
+ // 构建请求体：如果日志配置没有被修改，则不包含 log 字段
+ const requestBody: any = {
+ ssh: {
+ jumpHost: config.sshJumpHost,
+ jumpPort: config.sshJumpPort,
+ },
+ };
+
+ if (isLogConfigDirty) {
+ requestBody.log = config.log;
+ }
+
const resp = await fetch("/api/v1/system-config", {
method: "PUT",
headers: { "Content-Type": "application/json" },
- body: JSON.stringify(config),
+ body: JSON.stringify(requestBody),
});
if (!resp.ok) throw new Error(await resp.text());
setSaved(true);
@@ -92,8 +153,19 @@ export function SystemConfigPage({ copy: c }: { copy: Copy }) {
onClick={fetchConfig}
title={zh ? "刷新" : "Refresh"}
disabled={loading}
+ aria-busy={loading}
>
- {loading ? "…" : zh ? "刷新" : "Refresh"}
+
+ {loading
+ ? zh
+ ? "刷新中..."
+ : "Refreshing..."
+ : zh
+ ? "刷新"
+ : "Refresh"}
|