learning_ai_common_plat/dashboards/admin-web/src/lib/ops-stack.ts

import net from 'node:net';

export type ServiceStatus = 'healthy' | 'degraded' | 'down' | 'maintenance';
export type CheckKind = 'http-json' | 'http-status' | 'tcp';
export type InventorySource = 'docker' | 'vm';

export interface ServiceCheck {
  id: string;
  name: string;
  group: string;
  target: string;
  status: ServiceStatus;
  latency: number;
  version?: string;
  message?: string;
  lastChecked: string;
}

export interface OpsStatus {
  overall: 'healthy' | 'degraded' | 'critical';
  timestamp: string;
  services: ServiceCheck[];
}

interface BaseDefinition {
  id: string;
  name: string;
  group: string;
  description: string;
  management: InventorySource;
  exposure: 'internal' | 'public';
  port?: number;
}

interface HttpServiceDefinition extends BaseDefinition {
  kind: 'http-json' | 'http-status';
  env?: string;
  default: string;
  path: string;
}

interface TcpServiceDefinition extends BaseDefinition {
  kind: 'tcp';
  host: string;
  port: number;
}

export type ServiceDefinition = HttpServiceDefinition | TcpServiceDefinition;

export interface InventoryService extends ServiceCheck {
  description: string;
  management: InventorySource;
  exposure: 'internal' | 'public';
  port?: number;
  restartable: boolean;
}

export interface HostTool {
  id: string;
  name: string;
  group: string;
  source: InventorySource;
  management: string;
  status: 'managed' | 'manual';
  description: string;
}

export const STACK_SERVICES: ServiceDefinition[] = [
  {
    id: 'admin-web',
    name: 'Admin Dashboard',
    group: 'Dashboards',
    description: 'Internal admin portal for platform review and ops workflows.',
    management: 'docker',
    exposure: 'internal',
    port: 3001,
    kind: 'http-status',
    default: 'http://admin-web:3001',
    path: '/api/health',
  },
  {
    id: 'tracker-web',
    name: 'Tracker Dashboard',
    group: 'Dashboards',
    description: 'Internal tracker UI for issue and delivery review.',
    management: 'docker',
    exposure: 'internal',
    port: 3003,
    kind: 'http-status',
    default: 'http://tracker-web:3003',
    path: '/api/health',
  },
  {
    id: 'lysnrai-dashboard',
    name: 'LysnrAI Dashboard',
    group: 'Product Web Apps',
    description: 'Voice AI dashboard hosted on the VM for internal product review.',
    management: 'docker',
    exposure: 'internal',
    port: 3002,
    kind: 'http-status',
    default: 'http://lysnrai-dashboard:3002',
    path: '/',
  },
  {
    id: 'chronomind-web',
    name: 'ChronoMind Web',
    group: 'Product Web Apps',
    description: 'ChronoMind web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3030,
    kind: 'http-status',
    default: 'http://chronomind-web:3030',
    path: '/',
  },
  {
    id: 'jarvisjr-web',
    name: 'JarvisJr Web',
    group: 'Product Web Apps',
    description: 'JarvisJr web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3035,
    kind: 'http-status',
    default: 'http://jarvisjr-web:3035',
    path: '/',
  },
  {
    id: 'flowmonk-web',
    name: 'FlowMonk Web',
    group: 'Product Web Apps',
    description: 'FlowMonk web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3040,
    kind: 'http-status',
    default: 'http://flowmonk-web:3040',
    path: '/',
  },
  {
    id: 'notelett-web',
    name: 'NoteLett Web',
    group: 'Product Web Apps',
    description: 'NoteLett web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3045,
    kind: 'http-status',
    default: 'http://notelett-web:3045',
    path: '/',
  },
  {
    id: 'mindlyst-web',
    name: 'MindLyst Web',
    group: 'Product Web Apps',
    description: 'MindLyst web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3050,
    kind: 'http-status',
    default: 'http://mindlyst-web:3050',
    path: '/',
  },
  {
    id: 'nomgap-web',
    name: 'NomGap Web',
    group: 'Product Web Apps',
    description: 'NomGap web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3055,
    kind: 'http-status',
    default: 'http://nomgap-web:3055',
    path: '/',
  },
  {
    id: 'actiontrail-web',
    name: 'ActionTrail Web',
    group: 'Product Web Apps',
    description: 'ActionTrail web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3060,
    kind: 'http-status',
    default: 'http://actiontrail-web:3060',
    path: '/',
  },
  {
    id: 'llmlab-dashboard',
    name: 'LLM Lab Dashboard',
    group: 'Internal Tooling',
    description: 'Internal LLM lab dashboard hosted on the VM for operator and developer use.',
    management: 'docker',
    exposure: 'internal',
    port: 3075,
    kind: 'http-status',
    default: 'http://llmlab-dashboard:3075',
    path: '/',
  },
  {
    id: 'efforise-web',
    name: 'Efforise Web',
    group: 'Product Web Apps',
    description: 'Efforise web client hosted on the VM.',
    management: 'docker',
    exposure: 'internal',
    port: 3080,
    kind: 'http-status',
    default: 'http://efforise-web:3080',
    path: '/',
  },
  {
    id: 'platform',
    name: 'Platform Service',
    group: 'Core Services',
    description: 'Core API and auth platform service.',
    management: 'docker',
    exposure: 'internal',
    port: 4003,
    env: 'PLATFORM_SERVICE_URL',
    kind: 'http-json',
    default: 'http://platform-service:4003',
    path: '/health',
  },
  {
    id: 'extraction',
    name: 'Extraction Service',
    group: 'Core Services',
    description: 'Structured extraction service with product-aware throttling.',
    management: 'docker',
    exposure: 'internal',
    port: 4005,
    env: 'EXTRACTION_SERVICE_URL',
    kind: 'http-json',
    default: 'http://extraction-service:4005',
    path: '/health',
  },
  {
    id: 'mcp',
    name: 'MCP Server',
    group: 'Core Services',
    description: 'Internal MCP integration surface.',
    management: 'docker',
    exposure: 'internal',
    port: 4007,
    env: 'MCP_SERVER_URL',
    kind: 'http-json',
    default: 'http://mcp-server:4007',
    path: '/health',
  },
  {
    id: 'grafana',
    name: 'Grafana',
    group: 'Observability',
    description: 'Metrics and logs visualization.',
    management: 'docker',
    exposure: 'internal',
    port: 3000,
    kind: 'http-json',
    default: 'http://grafana:3000',
    path: '/api/health',
  },
  {
    id: 'loki',
    name: 'Loki',
    group: 'Observability',
    description: 'Centralized log aggregation.',
    management: 'docker',
    exposure: 'internal',
    port: 3100,
    kind: 'http-status',
    default: 'http://loki:3100',
    path: '/ready',
  },
  {
    id: 'prometheus',
    name: 'Prometheus',
    group: 'Observability',
    description: 'Internal metrics scraping and query engine.',
    management: 'docker',
    exposure: 'internal',
    port: 9090,
    kind: 'http-status',
    default: 'http://prometheus:9090',
    path: '/-/healthy',
  },
  {
    id: 'node-exporter',
    name: 'Node Exporter',
    group: 'Observability',
    description: 'Host-level VM metrics exporter.',
    management: 'docker',
    exposure: 'internal',
    port: 9100,
    kind: 'http-status',
    default: 'http://node-exporter:9100',
    path: '/metrics',
  },
  {
    id: 'cadvisor',
    name: 'cAdvisor',
    group: 'Observability',
    description: 'Container-level metrics exporter.',
    management: 'docker',
    exposure: 'internal',
    port: 8080,
    kind: 'http-status',
    default: 'http://cadvisor:8080',
    path: '/healthz',
  },
  {
    id: 'valkey',
    name: 'Valkey',
    group: 'Shared Infrastructure',
    description: 'Shared cache and rate-limit backing store.',
    management: 'docker',
    exposure: 'internal',
    kind: 'tcp',
    host: 'valkey',
    port: 6379,
  },
  {
    id: 'gitea-registry',
    name: 'Gitea Registry',
    group: 'Shared Infrastructure',
    description: 'Private npm package registry and source control service.',
    management: 'docker',
    exposure: 'internal',
    port: 3300,
    kind: 'http-json',
    default: 'http://gitea-npm-registry:3000',
    path: '/api/v1/version',
  },
  {
    id: 'mailpit',
    name: 'Mailpit',
    group: 'Shared Infrastructure',
    description: 'SMTP sink and email inspection UI.',
    management: 'docker',
    exposure: 'internal',
    port: 8025,
    kind: 'http-status',
    default: 'http://mailpit:8025',
    path: '/',
  },
  {
    id: 'azurite',
    name: 'Azurite',
    group: 'Shared Infrastructure',
    description: 'Local Azure Blob Storage emulator.',
    management: 'docker',
    exposure: 'internal',
    kind: 'tcp',
    host: 'azurite',
    port: 10000,
  },
  {
    id: 'cosmos-emulator',
    name: 'Cosmos Emulator',
    group: 'Shared Infrastructure',
    description: 'Local Azure Cosmos DB emulator.',
    management: 'docker',
    exposure: 'internal',
    port: 8080,
    kind: 'http-status',
    default: 'http://cosmos-emulator:8080',
    path: '/ready',
  },
  {
    id: 'gateway',
    name: 'Traefik Gateway',
    group: 'Ingress',
    description: 'Legacy internal gateway and routing layer.',
    management: 'docker',
    exposure: 'internal',
    port: 8080,
    kind: 'http-status',
    default: 'http://gateway:8080',
    path: '/',
  },
  {
    id: 'caddy',
    name: 'Caddy',
    group: 'Ingress',
    description: 'HTTPS ingress and reverse proxy for internal and backend domains.',
    management: 'docker',
    exposure: 'public',
    kind: 'tcp',
    host: 'caddy',
    port: 80,
  },
];

export const HOST_TOOLS: HostTool[] = [
  {
    id: 'docker-ce',
    name: 'Docker CE',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM bootstrap',
    status: 'managed',
    description: 'Container runtime for the internal stack.',
  },
  {
    id: 'docker-compose',
    name: 'Docker Compose',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM bootstrap',
    status: 'managed',
    description: 'Multi-service orchestration for the VM stack.',
  },
  {
    id: 'azure-cli',
    name: 'Azure CLI',
    group: 'Host Tooling',
    source: 'vm',
    management: 'Manual install',
    status: 'manual',
    description: 'Azure subscription and NSG management from the VM.',
  },
  {
    id: 'nodejs',
    name: 'Node.js 22',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM bootstrap',
    status: 'managed',
    description: 'Build/runtime toolchain for workspace services.',
  },
  {
    id: 'pnpm',
    name: 'pnpm',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM bootstrap',
    status: 'managed',
    description: 'Workspace package manager.',
  },
  {
    id: 'git',
    name: 'git',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM bootstrap',
    status: 'managed',
    description: 'Repo sync and deployment workflow tooling.',
  },
  {
    id: 'jq',
    name: 'jq',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM bootstrap',
    status: 'managed',
    description: 'CLI JSON inspection used in ops and setup scripts.',
  },
  {
    id: 'caddy-host-config',
    name: 'Caddy Config',
    group: 'Host Tooling',
    source: 'vm',
    management: 'VM file mount',
    status: 'managed',
    description: 'Host-mounted Caddy configuration at /opt/bytelyst/Caddyfile.',
  },
];

export const RESTARTABLE_SERVICE_CONTAINERS: Record<string, string> = {
  'admin-web': 'learning_ai_common_plat-admin-web-1',
  'tracker-web': 'learning_ai_common_plat-tracker-web-1',
  'lysnrai-dashboard': 'learning_ai_common_plat-lysnrai-dashboard-1',
  'chronomind-web': 'learning_ai_common_plat-chronomind-web-1',
  'jarvisjr-web': 'learning_ai_common_plat-jarvisjr-web-1',
  'flowmonk-web': 'learning_ai_common_plat-flowmonk-web-1',
  'notelett-web': 'learning_ai_common_plat-notelett-web-1',
  'mindlyst-web': 'learning_ai_common_plat-mindlyst-web-1',
  'nomgap-web': 'learning_ai_common_plat-nomgap-web-1',
  'actiontrail-web': 'learning_ai_common_plat-actiontrail-web-1',
  'llmlab-dashboard': 'learning_ai_common_plat-llmlab-dashboard-1',
  'efforise-web': 'learning_ai_common_plat-efforise-web-1',
  platform: 'learning_ai_common_plat-platform-service-1',
  extraction: 'learning_ai_common_plat-extraction-service-1',
  mcp: 'learning_ai_common_plat-mcp-server-1',
  grafana: 'learning_ai_common_plat-grafana-1',
  loki: 'learning_ai_common_plat-loki-1',
  prometheus: 'learning_ai_common_plat-prometheus-1',
  'node-exporter': 'learning_ai_common_plat-node-exporter-1',
  cadvisor: 'learning_ai_common_plat-cadvisor-1',
  valkey: 'learning_ai_common_plat-valkey-1',
  'gitea-registry': 'gitea-npm-registry',
  mailpit: 'learning_ai_common_plat-mailpit-1',
  azurite: 'learning_ai_common_plat-azurite-1',
  'cosmos-emulator': 'learning_ai_common_plat-cosmos-emulator-1',
};

async function checkHttpService(service: HttpServiceDefinition): Promise<ServiceCheck> {
  const baseUrl = (service.env && process.env[service.env]) || service.default;
  const target = `${baseUrl}${service.path}`;
  const start = Date.now();

  try {
    const res = await fetch(target, {
      method: 'GET',
      headers: { 'Content-Type': 'application/json' },
      next: { revalidate: 0 },
      signal: AbortSignal.timeout(3000),
    });

    const latency = Date.now() - start;

    if (!res.ok) {
      return {
        id: service.id,
        name: service.name,
        group: service.group,
        target,
        status: 'down',
        latency,
        message: `HTTP ${res.status}`,
        lastChecked: new Date().toISOString(),
      };
    }

    if (service.kind === 'http-json') {
      const payload = await res.json().catch(() => null);
      const rawStatus = payload?.status;
      const isOk =
        rawStatus === 'ok' ||
        rawStatus === 'healthy' ||
        payload?.database === 'ok' ||
        payload?.commit === 'ok' ||
        payload?.version;

      return {
        id: service.id,
        name: service.name,
        group: service.group,
        target,
        status: isOk ? 'healthy' : 'degraded',
        latency,
        version: payload?.version,
        message: isOk ? undefined : JSON.stringify(payload),
        lastChecked: new Date().toISOString(),
      };
    }

    return {
      id: service.id,
      name: service.name,
      group: service.group,
      target,
      status: 'healthy',
      latency,
      lastChecked: new Date().toISOString(),
    };
  } catch (err) {
    return {
      id: service.id,
      name: service.name,
      group: service.group,
      target,
      status: 'down',
      latency: Date.now() - start,
      message: err instanceof Error ? err.message : String(err),
      lastChecked: new Date().toISOString(),
    };
  }
}

async function checkTcpService(service: TcpServiceDefinition): Promise<ServiceCheck> {
  const start = Date.now();
  const target = `${service.host}:${service.port}`;

  return new Promise(resolve => {
    const socket = net.createConnection({ host: service.host, port: service.port });
    let settled = false;

    const finish = (status: ServiceStatus, message?: string) => {
      if (settled) return;
      settled = true;
      socket.destroy();
      resolve({
        id: service.id,
        name: service.name,
        group: service.group,
        target,
        status,
        latency: Date.now() - start,
        message,
        lastChecked: new Date().toISOString(),
      });
    };

    socket.setTimeout(3000);
    socket.once('connect', () => finish('healthy'));
    socket.once('timeout', () => finish('down', 'Connection timed out'));
    socket.once('error', err => finish('down', err.message));
  });
}

export async function collectOpsChecks(): Promise<ServiceCheck[]> {
  return Promise.all(
    STACK_SERVICES.map(service =>
      service.kind === 'tcp' ? checkTcpService(service) : checkHttpService(service)
    )
  );
}

export async function collectOpsStatus(): Promise<OpsStatus> {
  const services = await collectOpsChecks();
  const downCount = services.filter(c => c.status === 'down').length;
  const degradedCount = services.filter(c => c.status === 'degraded').length;

  let overall: OpsStatus['overall'] = 'healthy';
  if (downCount > 0) overall = 'critical';
  else if (degradedCount > 0) overall = 'degraded';

  return {
    overall,
    timestamp: new Date().toISOString(),
    services,
  };
}

export async function collectInventoryServices(): Promise<InventoryService[]> {
  const checks = await collectOpsChecks();
  const byId = new Map(checks.map(check => [check.id, check]));

  return STACK_SERVICES.map(service => {
    const check = byId.get(service.id);
    return {
      ...(check as ServiceCheck),
      description: service.description,
      management: service.management,
      exposure: service.exposure,
      port: service.port,
      restartable: Boolean(RESTARTABLE_SERVICE_CONTAINERS[service.id]),
    };
  });
}