feat(system): add Worker 006 telemetry and network profile

2026-07-27 21:06:51 +03:00
parent 438196cd6d
commit 67e12a47a4
17 changed files with 2723 additions and 8 deletions
@@ -46,6 +46,8 @@ import type { WorkspaceRendererProps } from "./contracts";
 import { DatasetGatewayWorkspace } from "./DatasetGatewayWorkspace";
 import { ContourHealthWorkspace } from "./ContourHealthWorkspace";
 import { LaboratoryArchiveWorkspace } from "./laboratory/LaboratoryArchiveWorkspace";
+import { ComputeModulesWorkspace } from "./system/ComputeModulesWorkspace";
+import { NetworkWorkspace } from "./system/NetworkWorkspace";

 function statusTone(status: CapabilityStatus): "success" | "accent" | "warning" | "neutral" {
  if (status === "active") return "success";
@@ -1177,6 +1179,10 @@ export function WorkspaceRenderer(props: WorkspaceRendererProps) {
          state={props.state}
        />
      );
+    case "compute-modules":
+      return <ComputeModulesWorkspace />;
+    case "network-monitor":
+      return <NetworkWorkspace />;
    case "datasets":
      return <DatasetGatewayWorkspace />;
    case "lab-archive":
@@ -0,0 +1,226 @@
+import {
+  Button,
+  GlassSurface,
+  Icon,
+  StatusBadge,
+} from "@nodedc/ui-react";
+
+import { TelemetrySeries } from "../../components/system/TelemetrySeries";
+import { WorkerRuntimeCard } from "../../components/system/WorkerRuntimeCard";
+import {
+  formatBytes,
+  formatDuration,
+  formatOptionalPercent,
+} from "../../components/system/systemFormat";
+import { useWorkerTelemetry } from "../../core/system/useWorkerTelemetry";
+
+function pipelineStateLabel(state: string): string {
+  if (state === "busy") return "Выполняет задачу";
+  if (state === "ready") return "Готов к задаче";
+  return "Нет live-состояния";
+}
+
+export function ComputeModulesWorkspace() {
+  const { telemetry, loading, error, refresh } = useWorkerTelemetry();
+  const node = telemetry?.node ?? null;
+  const missionCoreRuntimes = telemetry?.runtimes.filter((runtime) => !runtime.external) ?? [];
+  const externalRuntimes = telemetry?.runtimes.filter((runtime) => runtime.external) ?? [];
+  const connected = Boolean(
+    telemetry?.connection.reachable && telemetry.connection.identity_matches && node,
+  );
+  const history = telemetry?.history ?? [];
+  const cpuPercent = node?.cpu.load_percent;
+  const memoryPercent = node?.memory.used_percent;
+  const gpuPercent = node?.gpu?.utilization_percent;
+  const gpuMemoryPercent = node?.gpu?.memory_used_percent;
+
+  return (
+    <div className="system-workspace compute-modules-workspace">
+      <section className="system-workspace__lead">
+        <div>
+          <span className="section-eyebrow">СИСТЕМА / ВЫЧИСЛИТЕЛЬНЫЕ МОДУЛИ</span>
+          <h2>Worker 006</h2>
+          <p>
+            Живой аппаратный и процессинговый срез выделенного узла. Нагрузка внешних
+            сервисов отделена от Mission Core и не входит в оценку наших runtime.
+          </p>
+        </div>
+        <div className="system-workspace__actions">
+          <StatusBadge tone={connected ? "success" : "danger"}>
+            {connected ? "Узел доступен" : "Нет связи с узлом"}
+          </StatusBadge>
+          <Button
+            size="compact"
+            variant="secondary"
+            disabled={loading}
+            onClick={refresh}
+            icon={<Icon name="refresh" size={14} />}
+          >
+            {loading ? "Читаем" : "Обновить"}
+          </Button>
+        </div>
+      </section>
+
+      {error ? (
+        <GlassSurface className="system-workspace__notice" padding="md" tone="soft">
+          <StatusBadge tone="warning">{error}</StatusBadge>
+        </GlassSurface>
+      ) : null}
+
+      <section className="system-telemetry-grid" aria-label="Аппаратная телеметрия">
+        <TelemetrySeries
+          label="CPU"
+          value={formatOptionalPercent(cpuPercent)}
+          ceiling={100}
+          values={history.map((item) => item.cpu_percent)}
+        />
+        <TelemetrySeries
+          label="RAM"
+          value={formatOptionalPercent(memoryPercent)}
+          ceiling={100}
+          values={history.map((item) => item.memory_percent)}
+        />
+        <TelemetrySeries
+          label="GPU"
+          value={formatOptionalPercent(gpuPercent)}
+          ceiling={100}
+          values={history.map((item) => item.gpu_percent)}
+        />
+        <TelemetrySeries
+          label="VRAM"
+          value={formatOptionalPercent(gpuMemoryPercent)}
+          ceiling={100}
+          values={history.map((item) => item.gpu_memory_percent)}
+        />
+      </section>
+
+      <GlassSurface className="worker-hardware" padding="lg">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">HARDWARE</span>
+            <h3>{node?.node_id ?? telemetry?.profile.expected_node_id ?? "Worker 006"}</h3>
+            <p>{node?.os.caption ?? "Аппаратный профиль недоступен"}</p>
+          </div>
+          <StatusBadge tone={node?.node_id ? "success" : "danger"}>
+            {node?.node_id ? telemetry?.profile.display_name : "Нет данных"}
+          </StatusBadge>
+        </header>
+        <div className="worker-hardware__facts">
+          <dl>
+            <div><dt>Процессор</dt><dd>{node?.cpu.name ?? "—"}</dd></div>
+            <div><dt>Логические ядра</dt><dd>{node?.cpu.logical_processors ?? "—"}</dd></div>
+            <div><dt>Память занята</dt><dd>{formatBytes(node?.memory.used_bytes)}</dd></div>
+            <div><dt>Uptime</dt><dd>{formatDuration(node?.os.uptime_seconds)}</dd></div>
+          </dl>
+          <dl>
+            <div><dt>GPU</dt><dd>{node?.gpu?.name ?? "—"}</dd></div>
+            <div><dt>VRAM занята</dt><dd>{formatBytes(
+              typeof node?.gpu?.memory_used_mib === "number"
+                ? node.gpu.memory_used_mib * 1024 * 1024
+                : null,
+            )}</dd></div>
+            <div><dt>Температура</dt><dd>{
+              typeof node?.gpu?.temperature_celsius === "number"
+                ? `${node.gpu.temperature_celsius} °C`
+                : "—"
+            }</dd></div>
+            <div><dt>Мощность</dt><dd>{
+              typeof node?.gpu?.power_watts === "number"
+                ? `${node.gpu.power_watts.toFixed(1)} Вт`
+                : "—"
+            }</dd></div>
+          </dl>
+        </div>
+        <div className="worker-disk-list">
+          {(node?.disks ?? []).map((disk) => {
+            const size = typeof disk.size_bytes === "number" ? disk.size_bytes : null;
+            const free = typeof disk.free_bytes === "number" ? disk.free_bytes : null;
+            const used = size !== null && free !== null ? size - free : null;
+            return (
+              <div key={disk.name ?? "disk"}>
+                <span>Диск {disk.name ?? "—"}</span>
+                <strong>{formatBytes(used)} / {formatBytes(size)}</strong>
+              </div>
+            );
+          })}
+        </div>
+      </GlassSurface>
+
+      <section className="system-runtime-section">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">PROCESSING RUNTIME</span>
+            <h3>Контейнеры Mission Core</h3>
+            <p>Два ограниченных runtime: inference server и прикладной perception pipeline.</p>
+          </div>
+          <StatusBadge tone={node?.triton.ready ? "success" : "danger"}>
+            {node?.triton.ready ? "Triton ready" : "Triton недоступен"}
+          </StatusBadge>
+        </header>
+        <div className="worker-runtime-grid">
+          {missionCoreRuntimes.map((runtime) => (
+            <WorkerRuntimeCard key={runtime.name} runtime={runtime} />
+          ))}
+        </div>
+      </section>
+
+      <GlassSurface className="worker-pipeline" padding="lg">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">ТЕКУЩАЯ ЗАДАЧА</span>
+            <h3>{pipelineStateLabel(telemetry?.pipeline.service_state ?? "unavailable")}</h3>
+            <p>
+              {telemetry?.pipeline.active_request_id
+                ? `Run ${telemetry.pipeline.active_request_id}`
+                : "Очередь свободна; модели остаются загруженными в persistent worker."}
+            </p>
+          </div>
+          <StatusBadge tone={
+            telemetry?.pipeline.service_state === "busy" ? "warning"
+              : telemetry?.pipeline.service_state === "ready" ? "success"
+                : "danger"
+          }>
+            {telemetry?.pipeline.service_state ?? "unavailable"}
+          </StatusBadge>
+        </header>
+        <div className="worker-pipeline__summary">
+          <div><span>Завершено запусков</span><strong>{telemetry?.pipeline.completed_runs ?? "—"}</strong></div>
+          <div><span>Ошибок запусков</span><strong>{telemetry?.pipeline.failed_runs ?? "—"}</strong></div>
+          <div><span>Inference success</span><strong>{node?.triton.requests_succeeded ?? "—"}</strong></div>
+          <div><span>Inference failed</span><strong>{node?.triton.requests_failed ?? "—"}</strong></div>
+        </div>
+        <ol className="worker-stage-list">
+          {(telemetry?.pipeline.stages ?? []).map((stage, index) => (
+            <li key={stage.id} data-state={stage.state}>
+              <span>{String(index + 1).padStart(2, "0")}</span>
+              <strong>{stage.label}</strong>
+              <small>{
+                stage.state === "active" ? "выполняется"
+                  : stage.state === "waiting" ? "в очереди"
+                    : stage.state === "ready" ? "готов"
+                      : "нет данных"
+              }</small>
+            </li>
+          ))}
+        </ol>
+      </GlassSurface>
+
+      <section className="system-runtime-section system-runtime-section--external">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">CO-TENANTS / НЕ MISSION CORE</span>
+            <h3>Внешняя нагрузка узла</h3>
+            <p>
+              Эти процессы влияют на общую плату и VRAM, но не считаются расходом Mission Core.
+            </p>
+          </div>
+        </header>
+        <div className="worker-runtime-grid">
+          {externalRuntimes.map((runtime) => (
+            <WorkerRuntimeCard key={runtime.name} runtime={runtime} />
+          ))}
+        </div>
+      </section>
+    </div>
+  );
+}
@@ -0,0 +1,311 @@
+import { useEffect, useMemo, useState } from "react";
+import {
+  Button,
+  GlassSurface,
+  Icon,
+  StatusBadge,
+  TextField,
+} from "@nodedc/ui-react";
+
+import { TelemetrySeries } from "../../components/system/TelemetrySeries";
+import {
+  formatBitRate,
+  formatBytes,
+  formatLatency,
+  formatRate,
+} from "../../components/system/systemFormat";
+import {
+  fetchWorkerProfile,
+  saveWorkerProfile,
+  testWorkerProfile,
+  type WorkerConnectionProfile,
+  type WorkerProbe,
+} from "../../core/system/workerTelemetry";
+import { useWorkerTelemetry } from "../../core/system/useWorkerTelemetry";
+
+type ProfileAction = "idle" | "testing" | "saving";
+
+function probeLabel(probe: WorkerProbe | null): string {
+  if (!probe) return "Изменения не проверены";
+  if (!probe.reachable) return "Адрес не отвечает";
+  if (!probe.identity_matches) return "Ответил другой узел";
+  return "Worker 006 подтверждён";
+}
+
+export function NetworkWorkspace() {
+  const { telemetry, loading, error, refresh } = useWorkerTelemetry();
+  const [profile, setProfile] = useState<WorkerConnectionProfile | null>(null);
+  const [address, setAddress] = useState("");
+  const [port, setPort] = useState("22");
+  const [profileAction, setProfileAction] = useState<ProfileAction>("idle");
+  const [probe, setProbe] = useState<WorkerProbe | null>(null);
+  const [profileError, setProfileError] = useState<string | null>(null);
+  const aggregate = telemetry?.network.aggregate ?? null;
+  const connected = Boolean(
+    telemetry?.connection.reachable && telemetry.connection.identity_matches,
+  );
+
+  useEffect(() => {
+    const controller = new AbortController();
+    void fetchWorkerProfile(controller.signal)
+      .then((document) => {
+        if (controller.signal.aborted) return;
+        setProfile(document.profile);
+        setAddress(document.profile.address);
+        setPort(String(document.profile.port));
+        setProfileError(null);
+      })
+      .catch((reason: unknown) => {
+        if (controller.signal.aborted) return;
+        setProfileError(reason instanceof Error ? reason.message : "Профиль не прочитан.");
+      });
+    return () => controller.abort();
+  }, []);
+
+  const mutation = useMemo(() => {
+    const numericPort = Number(port);
+    if (!profile || !Number.isInteger(numericPort) || numericPort < 1 || numericPort > 65535) {
+      return null;
+    }
+    return {
+      revision: profile.revision,
+      address: address.trim(),
+      port: numericPort,
+    };
+  }, [address, port, profile]);
+
+  const runTest = () => {
+    if (!mutation) return;
+    const controller = new AbortController();
+    setProfileAction("testing");
+    setProfileError(null);
+    setProbe(null);
+    void testWorkerProfile(mutation, controller.signal)
+      .then(setProbe)
+      .catch((reason: unknown) => {
+        setProfileError(reason instanceof Error ? reason.message : "Проверка не выполнена.");
+      })
+      .finally(() => setProfileAction("idle"));
+  };
+
+  const saveProfile = () => {
+    if (!mutation) return;
+    const controller = new AbortController();
+    setProfileAction("saving");
+    setProfileError(null);
+    void saveWorkerProfile(mutation, controller.signal)
+      .then((document) => {
+        setProfile(document.profile);
+        setAddress(document.profile.address);
+        setPort(String(document.profile.port));
+        setProbe(document.verification);
+        refresh();
+      })
+      .catch((reason: unknown) => {
+        setProfileError(reason instanceof Error ? reason.message : "Профиль не сохранён.");
+      })
+      .finally(() => setProfileAction("idle"));
+  };
+
+  return (
+    <div className="system-workspace network-workspace">
+      <section className="system-workspace__lead">
+        <div>
+          <span className="section-eyebrow">СИСТЕМА / СЕТЬ</span>
+          <h2>Локальный вычислительный контур</h2>
+          <p>
+            Переносимый профиль связи между Mission Core и Worker 006. Адрес можно заменить
+            при переходе в другую локальную сеть; идентичность узла проверяется до сохранения.
+          </p>
+        </div>
+        <div className="system-workspace__actions">
+          <StatusBadge tone={connected ? "success" : "danger"}>
+            {connected ? "Маршрут доступен" : "Маршрут недоступен"}
+          </StatusBadge>
+          <Button
+            size="compact"
+            variant="secondary"
+            disabled={loading}
+            onClick={refresh}
+            icon={<Icon name="refresh" size={14} />}
+          >
+            {loading ? "Читаем" : "Обновить"}
+          </Button>
+        </div>
+      </section>
+
+      {(error || profileError) ? (
+        <GlassSurface className="system-workspace__notice" padding="md" tone="soft">
+          <StatusBadge tone="warning">{profileError ?? error}</StatusBadge>
+        </GlassSurface>
+      ) : null}
+
+      <section className="network-overview-grid" aria-label="Сетевая телеметрия">
+        <TelemetrySeries
+          label="Приём"
+          value={formatRate(aggregate?.receive_bytes_per_second)}
+          values={(telemetry?.history ?? []).map(
+            (item) => item.network_receive_bytes_per_second,
+          )}
+        />
+        <TelemetrySeries
+          label="Передача"
+          value={formatRate(aggregate?.send_bytes_per_second)}
+          values={(telemetry?.history ?? []).map(
+            (item) => item.network_send_bytes_per_second,
+          )}
+        />
+        <div className="network-stat-card">
+          <span>Сбор телеметрии</span>
+          <strong>{formatLatency(telemetry?.connection.latency_ms)}</strong>
+          <small>полный SSH probe Worker 006</small>
+        </div>
+        <div className="network-stat-card">
+          <span>Активные интерфейсы</span>
+          <strong>{telemetry?.network.interfaces.length ?? "—"}</strong>
+          <small>по Windows network counters</small>
+        </div>
+      </section>
+
+      <GlassSurface className="network-profile" padding="lg">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">ПРОФИЛЬ ПОДКЛЮЧЕНИЯ</span>
+            <h3>Worker 006</h3>
+            <p>
+              Реальный Node ID: <code>{profile?.expected_node_id ?? "DESKTOP-OPJ8J04"}</code>.
+              Пустой адрес использует закреплённый локальный SSH-профиль.
+            </p>
+          </div>
+          <StatusBadge tone={
+            probe?.reachable && probe.identity_matches ? "success"
+              : probe ? "danger"
+                : connected ? "success" : "neutral"
+          }>
+            {probe ? probeLabel(probe) : connected ? "Текущий профиль работает" : "Не проверено"}
+          </StatusBadge>
+        </header>
+        <div className="network-profile__form">
+          <TextField
+            label="Адрес Worker 006"
+            hint="IP или hostname"
+            value={address}
+            placeholder="из SSH-профиля mission-gpu"
+            autoComplete="off"
+            spellCheck={false}
+            onChange={(event) => {
+              setAddress(event.target.value);
+              setProbe(null);
+            }}
+          />
+          <TextField
+            label="SSH-порт"
+            hint="1–65535"
+            type="number"
+            min={1}
+            max={65535}
+            value={port}
+            onChange={(event) => {
+              setPort(event.target.value);
+              setProbe(null);
+            }}
+          />
+          <div className="network-profile__buttons">
+            <Button
+              variant="secondary"
+              disabled={!mutation || profileAction !== "idle"}
+              onClick={runTest}
+            >
+              {profileAction === "testing" ? "Проверяем" : "Проверить"}
+            </Button>
+            <Button
+              variant="primary"
+              disabled={!mutation || profileAction !== "idle"}
+              onClick={saveProfile}
+            >
+              {profileAction === "saving" ? "Сохраняем" : "Проверить и применить"}
+            </Button>
+          </div>
+        </div>
+        <dl className="network-profile__security">
+          <div><dt>Транспорт</dt><dd>SSH · key-only</dd></div>
+          <div><dt>Host key</dt><dd>strict · pinned</dd></div>
+          <div><dt>Профиль</dt><dd>{profile?.ssh_host_alias ?? "mission-gpu"}</dd></div>
+          <div><dt>Учётные данные</dt><dd>не доступны интерфейсу</dd></div>
+        </dl>
+      </GlassSurface>
+
+      <GlassSurface className="network-topology" padding="lg">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">ФАКТИЧЕСКИЙ МАРШРУТ</span>
+            <h3>Поток управления и обработки</h3>
+            <p>Схема собрана из текущего профиля и live health runtime, без demo-узлов.</p>
+          </div>
+        </header>
+        <div className="network-route">
+          <div data-state="online">
+            <span>Операторский UI</span>
+            <strong>Browser</strong>
+            <small>127.0.0.1:8000</small>
+          </div>
+          <i aria-hidden="true" data-state="online" />
+          <div data-state="online">
+            <span>Control Plane</span>
+            <strong>Mission Core</strong>
+            <small>локальный API</small>
+          </div>
+          <i aria-hidden="true" data-state={connected ? "online" : "offline"} />
+          <div data-state={connected ? "online" : "offline"}>
+            <span>Вычислительный узел</span>
+            <strong>Worker 006</strong>
+            <small>{profile?.address || profile?.ssh_host_alias || "mission-gpu"}:{profile?.port ?? 22}</small>
+          </div>
+          <i aria-hidden="true" data-state={
+            telemetry?.runtimes.some((runtime) => !runtime.external && runtime.state === "running")
+              ? "online" : "offline"
+          } />
+          <div data-state={
+            telemetry?.runtimes.some((runtime) => !runtime.external && runtime.state === "running")
+              ? "online" : "offline"
+          }>
+            <span>Runtime</span>
+            <strong>Triton + Pipeline</strong>
+            <small>внутренний Docker-контур</small>
+          </div>
+        </div>
+      </GlassSurface>
+
+      <section className="network-interface-section">
+        <header className="system-section-heading">
+          <div>
+            <span className="section-eyebrow">ИНТЕРФЕЙСЫ WORKER 006</span>
+            <h3>Адаптеры и счётчики</h3>
+            <p>Только активные интерфейсы, которые вернул сам узел.</p>
+          </div>
+        </header>
+        <div className="network-interface-list">
+          {(telemetry?.network.interfaces ?? []).map((adapter) => (
+            <GlassSurface className="network-interface-card" padding="md" key={adapter.name}>
+              <header>
+                <div>
+                  <strong>{adapter.name}</strong>
+                  <small>{adapter.description ?? "Описание недоступно"}</small>
+                </div>
+                <StatusBadge tone={adapter.status === "Up" ? "success" : "warning"}>
+                  {adapter.status ?? "unknown"}
+                </StatusBadge>
+              </header>
+              <dl>
+                <div><dt>Адреса</dt><dd>{adapter.addresses.join(", ") || "—"}</dd></div>
+                <div><dt>Link</dt><dd>{formatBitRate(adapter.link_speed_bps)}</dd></div>
+                <div><dt>Получено</dt><dd>{formatBytes(adapter.received_bytes)}</dd></div>
+                <div><dt>Отправлено</dt><dd>{formatBytes(adapter.sent_bytes)}</dd></div>
+              </dl>
+            </GlassSurface>
+          ))}
+        </div>
+      </section>
+    </div>
+  );
+}