wulf-pulse/app/api/status/workers/route.ts

/**
 * GET /api/status/workers
 *   Heartbeat snapshot for the three in-process workers:
 *     • analyzer  — analyzer_jobs
 *     • rmm       — rmm_executions
 *     • sync      — sync_schedules / sync_history (proxy for the scheduler)
 *
 *   For each: last activity timestamp, in-flight count, last-1h success/
 *   failure totals.  Cheap — just SELECT COUNT(*) FILTER queries. */

import { NextResponse } from 'next/server';
import { requireAuth } from '@/lib/auth-utils';
import postgresClient from '@/lib/services/postgres-client';

export interface WorkerSnapshot {
  name: string;
  lastActivity: string | null;
  inFlight: number;
  oneHour: { success: number; failure: number };
}

export async function GET() {
  const { error } = await requireAuth();
  if (error) return error;

  const [analyzerRes, rmmRes, syncRes] = await Promise.all([
    postgresClient.query<{
      last_activity: string | null;
      in_flight: string;
      ok_1h: string;
      fail_1h: string;
    }>(
      `SELECT
         GREATEST(MAX(queued_at), MAX(started_at), MAX(finished_at))::text         AS last_activity,
         COUNT(*) FILTER (WHERE status IN ('queued','fetching','triaging','itglue','analyzing','deep_review'))::text AS in_flight,
         COUNT(*) FILTER (WHERE status = 'complete' AND finished_at >= NOW() - INTERVAL '1 hour')::text AS ok_1h,
         COUNT(*) FILTER (WHERE status = 'failed'   AND finished_at >= NOW() - INTERVAL '1 hour')::text AS fail_1h
       FROM analyzer_jobs`,
    ),
    postgresClient.query<{
      last_activity: string | null;
      in_flight: string;
      ok_1h: string;
      fail_1h: string;
    }>(
      `SELECT
         GREATEST(MAX(queued_at), MAX(started_at), MAX(completed_at))::text       AS last_activity,
         COUNT(*) FILTER (WHERE status IN ('queued','running'))::text             AS in_flight,
         COUNT(*) FILTER (WHERE status = 'complete' AND completed_at >= NOW() - INTERVAL '1 hour')::text AS ok_1h,
         COUNT(*) FILTER (WHERE status IN ('failed','timeout') AND completed_at >= NOW() - INTERVAL '1 hour')::text AS fail_1h
       FROM rmm_executions`,
    ),
    postgresClient.query<{
      last_run: string | null;
      ok_1h: string;
      fail_1h: string;
    }>(
      `SELECT
         MAX(last_run)::text                                                       AS last_run,
         COUNT(*) FILTER (WHERE last_status = 'success' AND last_run >= NOW() - INTERVAL '1 hour')::text AS ok_1h,
         COUNT(*) FILTER (WHERE last_status = 'failed'  AND last_run >= NOW() - INTERVAL '1 hour')::text AS fail_1h
       FROM sync_schedules
       WHERE is_enabled = true`,
    ),
  ]);

  const a = analyzerRes.rows[0];
  const r = rmmRes.rows[0];
  const s = syncRes.rows[0];

  const workers: WorkerSnapshot[] = [
    {
      name: 'Analyzer',
      lastActivity: a?.last_activity ?? null,
      inFlight: parseInt(a?.in_flight ?? '0', 10),
      oneHour: {
        success: parseInt(a?.ok_1h ?? '0', 10),
        failure: parseInt(a?.fail_1h ?? '0', 10),
      },
    },
    {
      name: 'RMM Overshell',
      lastActivity: r?.last_activity ?? null,
      inFlight: parseInt(r?.in_flight ?? '0', 10),
      oneHour: {
        success: parseInt(r?.ok_1h ?? '0', 10),
        failure: parseInt(r?.fail_1h ?? '0', 10),
      },
    },
    {
      name: 'Sync scheduler',
      lastActivity: s?.last_run ?? null,
      inFlight: 0,
      oneHour: {
        success: parseInt(s?.ok_1h ?? '0', 10),
        failure: parseInt(s?.fail_1h ?? '0', 10),
      },
    },
  ];

  return NextResponse.json({ workers });
}