artifacts/api-server/src/index.ts

import app from "./app";
import { logger } from "./lib/logger";
import { ensureLoaded } from "./lib/research-data";
import { startJobPoller } from "./lib/job-tracker";
import { startLatencyAlertMonitor } from "./llm/latency-alert-monitor";
import {
  loadPersistedLatencySamples,
  startLatencySamplePersistence,
} from "./llm/sample-persistence";
import { config } from "./lib/config";
import { db } from "@workspace/db";
import { inviteCodes } from "@workspace/db";
import { sql } from "drizzle-orm";
import { seedToolGraphOnce } from "./lib/tool-graph-seed";
import { seedDrugClipOnce } from "./lib/drugclip/seed";
import {
  toolGraphEnabled,
  startDeprecationScheduler,
} from "./lib/tool-graph";
import { startEvolutionScheduler } from "./lib/evolution/scheduler";
import { applyCapabilityMigrations } from "./lib/capability/migrations/apply.js";
import { loadRegisteredCapabilities } from "./lib/capability/loader.js";

// Production sanity check: if neither the env-derived invite list nor any
// row in `invite_codes` is present, brand-new sign-ups will all 403. Log a
// loud warning so operators notice before users do. We do not fail boot —
// the seed step (`pnpm --filter @workspace/api-server seed:prod`) is the
// supported recovery path and may be run after the server is up.
if (process.env["NODE_ENV"] === "production" && config.inviteCodes.length === 0) {
  void (async () => {
    try {
      const rows = await db
        .select({ n: sql<number>`count(*)::int` })
        .from(inviteCodes);
      const count = rows[0]?.n ?? 0;
      if (count === 0) {
        logger.error(
          "AUTH BOOTSTRAP: no invite codes available (INVITE_CODES env is empty AND invite_codes table is empty). " +
            "All registrations will return 403. Run `pnpm --filter @workspace/api-server seed:prod` " +
            "with INITIAL_INVITE_CODES set, or populate the INVITE_CODES env var, then redeploy.",
        );
      }
    } catch (err) {
      logger.warn(
        { err },
        "AUTH BOOTSTRAP: could not query invite_codes table to verify auth bootstrap state.",
      );
    }
  })();
}

// Warm research data cache in background after server starts.
void ensureLoaded().catch((err) =>
  logger.error({ err }, "research data warm-load failed"),
);

// Idempotently seed the cross-process tool capability graph (Task #147).
// Safe to call on every boot — the seed importer no-ops if the curated
// nodes are already present and verified.
if (toolGraphEnabled() && process.env["NODE_ENV"] !== "test") {
  void seedToolGraphOnce().catch((err) =>
    logger.warn({ err }, "tool-graph seed failed (continuing without graph)"),
  );
}

// Seed first-class tool networks (Task #176, Wave A). DrugCLIP is the
// inaugural network; future networks register here too. Idempotent.
if (process.env["NODE_ENV"] !== "test") {
  void seedDrugClipOnce().catch((err) =>
    logger.warn({ err }, "drugclip seed failed (continuing without network)"),
  );
}

// Begin polling the research engine for in-flight async jobs so we can
// broadcast completion events and post the proactive follow-up message.
startJobPoller();

// Watch provider-gateway latency and broadcast an announcement when a
// provider's p95 has been near its timeout ceiling for several minutes.
// Skip during tests so suites don't have to teardown a global timer.
if (process.env["NODE_ENV"] !== "test") {
  startLatencyAlertMonitor();
  // Restore the persisted recent-samples window so the admin sparkline
  // (Task #127) is populated immediately after a restart instead of
  // waiting on fresh traffic, then start the debounced flush loop that
  // mirrors new in-memory samples back to Postgres.
  void loadPersistedLatencySamples()
    .catch((err) =>
      logger.warn({ err }, "loadPersistedLatencySamples failed"),
    )
    .finally(() => {
      startLatencySamplePersistence();
    });
}

const rawPort = process.env["PORT"];

if (!rawPort) {
  throw new Error(
    "PORT environment variable is required but was not provided.",
  );
}

const port = Number(rawPort);

if (Number.isNaN(port) || port <= 0) {
  throw new Error(`Invalid PORT value: "${rawPort}"`);
}

const server = app.listen(port, () => {
  logger.info({ port }, "Server listening");
  // Task #255 / B2 — 启动时打印 7 类外部知识适配器状态(real / interface_only)。
  void import("./lib/capability/external-knowledge/index.ts").then((m) =>
    m.logAdapterStatusBanner({ info: (msg) => logger.info(msg) }),
  );
  // Task #242 / B1 — 把 capability schema 拉到当前版本(idempotent),
  // 然后扫 _registered/*.manifest.{ts,js,mjs} 自动注册 capability。
  // 顺序很重要:loader 里的 manifest 可能直接 insert capability 行,
  // 必须先确保 capabilities / capability_lifecycle_events 表存在。
  void applyCapabilityMigrations()
    .then(({ applied, skipped }) =>
      logger.info(
        { applied, skipped },
        "capability schema migrations applied",
      ),
    )
    .then(() => loadRegisteredCapabilities())
    .then(({ loaded, errors }) => {
      if (loaded.length > 0 || errors.length > 0) {
        logger.info(
          { loaded, errors },
          "capability registry boot scan complete",
        );
      } else {
        logger.info(
          "capability registry boot scan complete (0 manifests in _registered/)",
        );
      }
    })
    .catch((err) =>
      logger.error(
        { err },
        "capability boot pipeline failed (server still up; capability layer degraded)",
      ),
    );
});

// Tool-graph deprecation scheduler (#159). No-op unless
// DEPRECATION_DETECTOR_ENABLED=1; safe to leave enabled in production.
if (toolGraphEnabled()) {
  startDeprecationScheduler();
}

// Online-evolution flywheel cadence worker (Task #181). Calls
// evaluateTriggers + attemptAutoPromote + tickRollbackWatch on a fixed
// schedule so the loop runs without an operator hitting the admin
// page. Honours EVOLUTION_SCHEDULER_DISABLED=1 and is inert in tests.
if (process.env["NODE_ENV"] !== "test") {
  startEvolutionScheduler();
}

server.on("error", (err: NodeJS.ErrnoException) => {
  if (err.code === "EADDRINUSE") {
    logger.fatal(
      { err, port },
      `Port ${port} is already in use. Another process is bound to it; refusing to start.`,
    );
  } else if (err.code === "EACCES") {
    logger.fatal(
      { err, port },
      `Permission denied binding to port ${port}.`,
    );
  } else {
    logger.fatal({ err, port }, "Fatal error while starting HTTP server");
  }
  process.exit(1);
});