Method start

server/src/http.ts:8511–8672

(port: number = 3000)

  }

  async start(port: number = 3000): Promise<void> {
    // Initialize OpenTelemetry logging for PostHog (all log levels)
    const { initOtelLogs, emitLog } = await import('./utils/otel-logs.js');
    const { setLogHook } = await import('./logger.js');
    if (initOtelLogs()) {
      setLogHook(emitLog);
    }

    // Initialize PostHog error tracking (captures all logger.error() calls as exceptions)
    const { initPostHogErrorTracking } = await import('./utils/posthog.js');
    initPostHogErrorTracking();

    // Initialize database
    const { initializeDatabase, onPoolError } = await import("./db/client.js");
    const { runMigrations } = await import("./db/migrate.js");
    const { getDatabaseConfig } = await import("./config.js");
    const dbConfig = getDatabaseConfig();
    if (!dbConfig) {
      throw new Error("DATABASE_URL or DATABASE_PRIVATE_URL environment variable is required");
    }
    initializeDatabase(dbConfig);

    // Escalate pool-level errors to Slack
    onPoolError(() => {
      notifySystemError({ source: 'database-pool', errorMessage: 'Database pool error — check application logs' });
    });

    await runMigrations();

    // Validate the idempotency backend can actually query its table — fails
    // fast on a stale pool, missing migration, or wrong-credentials boot
    // rather than silently passing every mutating call to a broken backend.
    // No-ops when the store falls back to memoryBackend.
    //
    // Bounded with a 10s deadline because the pg pool has connectionTimeoutMillis=5000
    // but no statement_timeout — without this race, a hung query (e.g., DB starting
    // up, replica failover) would stall boot indefinitely and starve Fly's TCP healthcheck.
    const { getIdempotencyStore } = await import("./training-agent/idempotency.js");
    const probe = getIdempotencyStore().probe?.();
    if (probe) {
      await Promise.race([
        probe,
        new Promise<never>((_, reject) =>
          setTimeout(
            () => reject(new Error("Idempotency backend probe timed out after 10s — check DATABASE_URL and pool reachability")),
            10_000,
          ),
        ),
      ]);
    }

    // Sync organizations from WorkOS and Stripe to local database (dev environment support)
    if (AUTH_ENABLED && workos) {
      const orgDb = new OrganizationDatabase();

      // Sync WorkOS organizations first
      try {
        const result = await orgDb.syncFromWorkOS(workos);