char-slop/berry

git clone https://git.t4t.associates/char-slop/berry

Charlotte Somexperiment: faster shutdowned34ebd2026-10-22

import { homedir } from "node:os";
import { serveDir } from "@std/http/file-server";
import { fromFileUrl, join, resolve } from "@std/path";
import { builtinModels } from "@earendil-works/pi-ai/providers/all";
import { openNodeSqliteStorage } from "@earendil-works/pi-durable/storage/sqlite/node";
import { piAuthStore } from "./auth.ts";
import { type Host, openBerry } from "./berry.ts";
import { createHandler } from "./http.ts";
import { denoCheck, extensionLoader } from "./loader.ts";
import { RESTART } from "./supervisor.ts";
import { extensionWeb } from "./web.ts";

const token = Deno.env.get("BERRY_TOKEN");
if (!token) {
  console.error("BERRY_TOKEN must be set");
  Deno.exit(1);
}

const root = fromFileUrl(new URL("../", import.meta.url));
const directories = [
  join(root, "extensions"),
  resolve(Deno.env.get("XDG_CONFIG_HOME") || join(homedir(), ".config"), "berry", "extensions"),
];
const credentials = piAuthStore(Deno.env.get("BERRY_AUTH"));
const models = builtinModels({ credentials });
if (Deno.env.get("BERRY_DEV_MODEL")) (await import("./dev.ts")).installDevModel(models);
const storage = await openNodeSqliteStorage(
  Deno.env.get("BERRY_DB") ?? join(root, "data", "berry.sqlite"),
);

// a restart or a signal takes the process down; a check still running would hold it up
const checks = new AbortController();

const host: Host = {
  root,
  extensionDirectories: directories,
  // the server, and the extensions it loads at startup, which it imports by name rather than statically
  check: async () => {
    const started = performance.now();
    const failures = await denoCheck(
      root,
      [
        "server/main.ts",
        ...loader.files(),
      ],
      checks.signal,
    );
    // a restart's whole budget: this, then the stop below, then the new process saying it is listening
    console.log(`[berry] checked in ${Math.round(performance.now() - started)}ms`);
    return failures;
  },
  restart: () => void stop(RESTART),
  extensions: () => loader.entries(),
  reload: (name) => loader.reload(name),
};

const berry = await openBerry(storage, models, host);
const web = extensionWeb({
  cacheDirectory: join(
    Deno.env.get("XDG_CACHE_HOME") || join(homedir(), ".cache"),
    "berry",
    "frontends",
  ),
});
const loader = extensionLoader({ berry, web, root, directories });
await loader.loadAll();
loader.watch();
berry.harness.resume();

const handler = createHandler({
  berry,
  models,
  credentials,
  web,
  loader,
  host,
  token,
  // the built frontend, so one port serves everything; `deno task web` serves it with hmr instead
  fallback: (request) => serveDir(request, { fsRoot: join(root, "web", "dist"), quiet: true }),
});
const server = Deno.serve({
  hostname: Deno.env.get("BERRY_HOST") ?? "127.0.0.1",
  port: Number(Deno.env.get("BERRY_PORT") ?? 8787),
}, handler);
// from process start, so a restart's budget is this and the checked/stopped lines beside it
console.log(`[berry] ready in ${Math.round(performance.now())}ms`);

// once, whatever asks first: a signal or a restart; another request waits for that one
let stopping: Promise<never> | undefined;
/**
 * How long teardown may take before the process leaves anyway: the database is a WAL, and every child watches its own
 * stdin, so what is still closing is left to end on its own rather than holding a restart.
 */
const STOP_DEADLINE_MS = 2_000;
/** How long an in-flight request may take to finish before the process leaves without it. */
const REQUEST_GRACE_MS = 500;
function stop(code: number): Promise<never> {
  stopping ??= (async () => {
    checks.abort(new Error("berry is stopping"));
    const started = performance.now();
    // where a slow shutdown goes; the only place these numbers can be seen
    const phases: string[] = [];
    const phase = async (name: string, run: () => unknown) => {
      const at = performance.now();
      await run();
      phases.push(`${name} ${Math.round(performance.now() - at)}ms`);
    };
    const closing = (async () => {
      await phase("sockets", () => handler.closeSockets());
      // a request of its own, such as a model refresh from a slow provider, ends with the process rather than holding it
      await phase(
        "http",
        () =>
          Promise.race([
            server.shutdown(),
            new Promise((resolve) => setTimeout(resolve, REQUEST_GRACE_MS)),
          ]),
      );
      await phase("extensions", () => loader.close());
      // signals every running call, and waits for them to end; durable resumes the work when berry opens again
      await phase("berry", () => berry.close());
    })().catch((error) => console.error("[berry] stopping:", error));
    const overdue = await Promise.race([
      closing.then(() => false),
      new Promise<boolean>((resolve) => setTimeout(() => resolve(true), STOP_DEADLINE_MS)),
    ]);
    const left = ["sockets", "http", "extensions", "berry"].filter((name) =>
      !phases.some((line) => line.startsWith(`${name} `))
    );
    console.log(
      `[berry] stopped in ${Math.round(performance.now() - started)}ms (${phases.join(", ")})` +
        (overdue && left.length
          ? `; gave up on ${left.join(", ")} after ${STOP_DEADLINE_MS}ms`
          : ""),
    );
    Deno.exit(code);
  })();
  return stopping;
}
for (const signal of ["SIGINT", "SIGTERM"] as const) {
  Deno.addSignalListener(signal, () => void stop(0));
}