hub-runner 0.2.6 → 0.2.7

package/README.md

@@ -12,7 +12,7 @@ hub-runner start
 
 | Command | What it does |
 |---|---|
-| `hub-runner start [--take-over]` | Starts the runner under launchd (`~/Library/LaunchAgents/dev.hub.runner.plist`). It starts again at every login and is not restarted after a crash: run `start` again. `--take-over` claims the Mac slot from another Mac. |
+| `hub-runner start [--take-over]` | Starts the runner under launchd (`~/Library/LaunchAgents/dev.hub.runner.plist`). It starts again at every login and is not restarted after a crash: run `start` again. `--take-over` claims the Mac slot from another Mac on that start only: a later start at login claims without it. |
 | `hub-runner stop` | Stops it. |
 | `hub-runner restart` | Lets the open turns finish, then starts it again. |
 | `hub-runner status` | The runner's state, pid, version, bots, and whether launchd has it loaded. Exits 1 when it isn't running. |
@@ -33,5 +33,5 @@ npm i -g hub-runner@latest --min-release-age=0 && hub-runner restart
 
 - Node 24 or later, and `pnpm` on your PATH when you run `start`: the runner uses both to set the bots up.
 - The hub is `https://hosted-hub.playground-vercel.tools`; set `HUB_URL` when you run `login` for another one.
-- The HubHosted app only shows the runner's state. Quitting, signing out of or replacing the app doesn't touch the runner.
+- The HubHosted app runs a runner of its own for each workspace with a Mac bot, in `~/.hub/local/<scope>` under the launchd job `dev.hub.runner.<scope>`; this CLI acts on one of those when `HUB_RUNNER_DIR` and `HUB_RUNNER_LABEL` name it. An npm install keeps `~/.hub/local` and `dev.hub.runner`. Quitting, signing out of or replacing the app doesn't stop a runner.
 - If the runner doesn't start, `hub-runner status` and `hub-runner logs` say why.

package/check.mjs

@@ -196,18 +196,19 @@ async function stopTree(pgid, tree, { name, timeoutMs, warn }) {
   killTree(pgid, tree);
   return untilTreeGone(pgid, tree, KILL_GRACE_MS);
 }
-function ownsLeader(record, rows, signature) {
+var signature = (name, port) => ` dev --agent ${name} --port ${port} `;
+function ownsLeader(record, rows, signature2) {
   if (!(record.pid > 1) || record.pgid !== record.pid) return false;
   const row = rows.find((r) => r.pid === record.pid);
   if (!row || row.pgid !== record.pgid) return false;
-  if (!` ${row.command} `.includes(signature)) return false;
+  if (!` ${row.command} `.includes(signature2)) return false;
   const member = record.tree?.find((m) => m.pid === record.pid);
   if (member) return member.startedAt === row.startedAt;
   const recorded = Date.parse(record.startedAt);
   return !(Number.isFinite(row.startedAt) && Number.isFinite(recorded) && Math.abs(row.startedAt - recorded) > 5e3);
 }
-function ourTree(record, rows, signature) {
-  const leader = ownsLeader(record, rows, signature);
+function ourTree(record, rows, signature2) {
+  const leader = ownsLeader(record, rows, signature2);
   return { leader, tree: merge(leader ? treeOf(record.pid, rows) : [], liveMembers(record.tree, rows)) };
 }
 
@@ -255,7 +256,6 @@ function createSupervisor({
   const entries = /* @__PURE__ */ new Map();
   let closed = false;
   let starting = Promise.resolve();
-  const signature = (name, port) => ` dev --agent ${name} --port ${port} `;
   const argvFor = (name, port) => [...command, "dev", "--agent", name, "--port", String(port), "--no-ui", "--no-default-extensions", "--resume"];
   const envFor = (agent) => typeof env === "function" ? env(agent) : env;
   const fingerprintFor = (agent, spawnEnv = envFor(agent)) => fingerprint ? fingerprint({ agent, env: spawnEnv, argv: argvFor(agent.name, agent.port) }) : null;

package/lib/launchd.mjs

@@ -1,9 +1,9 @@
 // The runner's launchd job: a plist in the user's LaunchAgents, loaded in the GUI domain, so it starts at login and
 // outlives every app. `RunAtLoad` without `KeepAlive`: a crash waits for `hub-runner start`.
 import { execFileSync, spawnSync } from 'node:child_process';
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
+import { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
 import { dirname, join } from 'node:path';
-import { LABEL, appHome, credentialsFile, logFile, mainFile, plistFile, runnerDir, version } from './paths.mjs';
+import { LABEL, appHome, credentialsFile, logFile, mainFile, plistFile, runnerDir, takeOverFile, version } from './paths.mjs';
 import { deviceName } from './device.mjs';
 
 const xml = (s) => String(s).replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;');
@@ -41,9 +41,9 @@ function program(home) {
   return { node: process.execPath, main: mainFile(), pnpm, path: process.env.PATH, build: version() };
 }
 
-export function plist({ takeOver }) {
+export function plist() {
   const run = program(appHome());
-  const env = { HUB_PNPM: run.pnpm, PATH: run.path, HUB_RUNNER_DIR: runnerDir(), HUB_RUNNER_BUILD: run.build, HUB_DEVICE_NAME: deviceName(), ...(takeOver ? { HUB_RUNNER_TAKE_OVER: '1' } : {}) };
+  const env = { HUB_PNPM: run.pnpm, PATH: run.path, HUB_RUNNER_DIR: runnerDir(), HUB_RUNNER_BUILD: run.build, HUB_DEVICE_NAME: deviceName() };
   return `<?xml version="1.0" encoding="UTF-8"?>
 <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
 <plist version="1.0">
@@ -78,7 +78,10 @@ export function start({ takeOver = false } = {}) {
   if (!existsSync(credentialsFile())) throw new Error('not signed in: run `hub-runner login` first');
   mkdirSync(runnerDir(), { recursive: true });
   mkdirSync(dirname(plistFile()), { recursive: true });
-  writeFileSync(plistFile(), plist({ takeOver }));
+  writeFileSync(plistFile(), plist());
+  // A takeover is for the start it's asked on, never a later load of the plist (a login): the runner reads and deletes this.
+  if (takeOver) writeFileSync(takeOverFile(), '');
+  else rmSync(takeOverFile(), { force: true });
   if (loaded()) bootout();
   launchctl(['bootstrap', domain(), plistFile()]);
 }

package/lib/paths.mjs

@@ -2,11 +2,14 @@ import { readFileSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { join } from 'node:path';
 
-export const LABEL = 'dev.hub.runner';
+// The Hub app runs one runner per workspace, each under its own label (RunnerInstall.swift); an npm install keeps the one.
+export const LABEL = process.env.HUB_RUNNER_LABEL || 'dev.hub.runner';
 export const DEFAULT_HUB = 'https://hosted-hub.playground-vercel.tools';
 export const runnerDir = () => process.env.HUB_RUNNER_DIR || join(homedir(), '.hub', 'local');
 export const credentialsFile = () => join(runnerDir(), 'credentials.json');
 export const logFile = () => join(runnerDir(), 'runner.log');
+// `start --take-over` asks the next runner to start to take the claim over; it deletes the file (hosted-hub/runner/main.ts).
+export const takeOverFile = () => join(runnerDir(), 'take-over');
 export const plistFile = () => join(homedir(), 'Library', 'LaunchAgents', `${LABEL}.plist`);
 export const mainFile = () => new URL('../main.mjs', import.meta.url).pathname;
 export const version = () => JSON.parse(readFileSync(new URL('../package.json', import.meta.url), 'utf8')).version;

package/main.mjs

@@ -1580,27 +1580,27 @@ var require_adapter = __commonJS({
 var require_proper_lockfile = __commonJS({
   "../../hosted-hub/node_modules/proper-lockfile/index.js"(exports, module) {
     "use strict";
-    var lockfile2 = require_lockfile();
+    var lockfile3 = require_lockfile();
     var { toPromise, toSync, toSyncOptions } = require_adapter();
     async function lock(file, options) {
-      const release2 = await toPromise(lockfile2.lock)(file, options);
+      const release2 = await toPromise(lockfile3.lock)(file, options);
       return toPromise(release2);
     }
     function lockSync(file, options) {
-      const release2 = toSync(lockfile2.lock)(file, toSyncOptions(options));
+      const release2 = toSync(lockfile3.lock)(file, toSyncOptions(options));
       return toSync(release2);
     }
     function unlock(file, options) {
-      return toPromise(lockfile2.unlock)(file, options);
+      return toPromise(lockfile3.unlock)(file, options);
     }
     function unlockSync(file, options) {
-      return toSync(lockfile2.unlock)(file, toSyncOptions(options));
+      return toSync(lockfile3.unlock)(file, toSyncOptions(options));
     }
     function check(file, options) {
-      return toPromise(lockfile2.check)(file, options);
+      return toPromise(lockfile3.check)(file, options);
     }
     function checkSync(file, options) {
-      return toSync(lockfile2.check)(file, toSyncOptions(options));
+      return toSync(lockfile3.check)(file, toSyncOptions(options));
     }
     module.exports = lock;
     module.exports.lock = lock;
@@ -1613,11 +1613,11 @@ var require_proper_lockfile = __commonJS({
 });
 
 // ../../hosted-hub/runner/main.ts
-var import_proper_lockfile = __toESM(require_proper_lockfile(), 1);
+var import_proper_lockfile2 = __toESM(require_proper_lockfile(), 1);
 import { homedir as homedir4 } from "node:os";
 import { randomUUID } from "node:crypto";
-import { existsSync as existsSync11, mkdirSync as mkdirSync7, readFileSync as readFileSync10, realpathSync as realpathSync4, writeFileSync as writeFileSync7 } from "node:fs";
-import { join as join16 } from "node:path";
+import { existsSync as existsSync11, mkdirSync as mkdirSync8, readFileSync as readFileSync11, realpathSync as realpathSync4, rmSync as rmSync5, writeFileSync as writeFileSync7 } from "node:fs";
+import { join as join17 } from "node:path";
 
 // ../../hosted-hub/lib/clean-env.mjs
 var INHERITED_ENV = /^(EVE_|WORKFLOW_)/;
@@ -1861,18 +1861,19 @@ async function stopTree(pgid, tree, { name, timeoutMs, warn }) {
   killTree(pgid, tree);
   return untilTreeGone(pgid, tree, KILL_GRACE_MS);
 }
-function ownsLeader(record, rows, signature) {
+var signature = (name, port) => ` dev --agent ${name} --port ${port} `;
+function ownsLeader(record, rows, signature2) {
   if (!(record.pid > 1) || record.pgid !== record.pid) return false;
   const row = rows.find((r) => r.pid === record.pid);
   if (!row || row.pgid !== record.pgid) return false;
-  if (!` ${row.command} `.includes(signature)) return false;
+  if (!` ${row.command} `.includes(signature2)) return false;
   const member = record.tree?.find((m) => m.pid === record.pid);
   if (member) return member.startedAt === row.startedAt;
   const recorded = Date.parse(record.startedAt);
   return !(Number.isFinite(row.startedAt) && Number.isFinite(recorded) && Math.abs(row.startedAt - recorded) > 5e3);
 }
-function ourTree(record, rows, signature) {
-  const leader = ownsLeader(record, rows, signature);
+function ourTree(record, rows, signature2) {
+  const leader = ownsLeader(record, rows, signature2);
   return { leader, tree: merge(leader ? treeOf(record.pid, rows) : [], liveMembers(record.tree, rows)) };
 }
 
@@ -1920,7 +1921,6 @@ function createSupervisor({
   const entries = /* @__PURE__ */ new Map();
   let closed = false;
   let starting = Promise.resolve();
-  const signature = (name, port) => ` dev --agent ${name} --port ${port} `;
   const argvFor = (name, port) => [...command, "dev", "--agent", name, "--port", String(port), "--no-ui", "--no-default-extensions", "--resume"];
   const envFor = (agent) => typeof env === "function" ? env(agent) : env;
   const fingerprintFor = (agent, spawnEnv = envFor(agent)) => fingerprint ? fingerprint({ agent, env: spawnEnv, argv: argvFor(agent.name, agent.port) }) : null;
@@ -4552,8 +4552,79 @@ async function stopWithin(stop, ms, signal) {
   ]);
 }
 
+// ../../hosted-hub/runner/ports.ts
+var import_proper_lockfile = __toESM(require_proper_lockfile(), 1);
+import { mkdirSync as mkdirSync7, readFileSync as readFileSync10 } from "node:fs";
+import { join as join16 } from "node:path";
+var listening = async (port) => (await listeners(port)).length > 0;
+async function ownedPorts(stateDir, root) {
+  let records;
+  try {
+    records = JSON.parse(readFileSync10(join16(stateDir, "processes.json"), "utf8"));
+  } catch {
+    return /* @__PURE__ */ new Map();
+  }
+  const owned = /* @__PURE__ */ new Map();
+  const rows = table();
+  for (const r of Array.isArray(records) ? records : []) {
+    if (typeof r?.agent !== "string" || typeof r?.port !== "number" || r.workspace && r.workspace !== root) continue;
+    const { tree } = ourTree(r, rows, signature(r.agent, r.port));
+    const holders = await listeners(r.port).catch(() => []);
+    if (holders.length && holders.every((pid) => tree.some((m) => m.pid === pid))) owned.set(r.agent, r.port);
+  }
+  return owned;
+}
+function portClaims(dir2, warn) {
+  const held = /* @__PURE__ */ new Map();
+  mkdirSync7(dir2, { recursive: true });
+  const release2 = async (port) => {
+    const r = held.get(port);
+    held.delete(port);
+    await r?.().catch(() => {
+    });
+  };
+  return {
+    claim: async (port) => {
+      if (held.has(port)) return true;
+      const path = join16(dir2, `${port}.lock`);
+      const done = await import_proper_lockfile.default.lock(path, {
+        lockfilePath: path,
+        realpath: false,
+        stale: 1e4,
+        onCompromised: (e) => {
+          held.delete(port);
+          warn(`port ${port}: claim lost, ${e.message}`);
+        }
+      }).catch((error) => {
+        if (error.code !== "ELOCKED") throw error;
+        return null;
+      });
+      if (done) held.set(port, done);
+      return !!done;
+    },
+    release: release2,
+    releaseAll: () => Promise.all([...held.keys()].map(release2))
+  };
+}
+async function pickPorts(start, roster, taken, portBase, owned, portBusy, claims) {
+  const picked = [];
+  for (const name of start) {
+    const own = owned.get(name);
+    let port = own ?? portBase + roster.indexOf(name);
+    for (; ; port++) {
+      if (taken.has(port) || !await claims.claim(port)) continue;
+      if (port === own || !await portBusy(port)) break;
+      await claims.release(port);
+    }
+    taken.add(port);
+    picked.push({ name, port });
+  }
+  return picked;
+}
+
 // ../../hosted-hub/runner/hub-runner.ts
-var realBots = { stage: stageLocal, supervise: createSupervisor };
+var supervise = (o) => Object.assign(createSupervisor(o), { owned: () => ownedPorts(o.stateDir, o.root) });
+var realBots = { stage: stageLocal, supervise };
 var HeldElsewhere = class extends Error {
   holder;
   constructor(holder) {
@@ -4581,6 +4652,7 @@ function botEnv(env, config, bot, token) {
 var log = (...args) => console.error((/* @__PURE__ */ new Date()).toISOString(), "[hub-runner]", ...args);
 async function run3(config, bots = realBots) {
   const shutdown = new AbortController();
+  const portBusy = config.portBusy ?? listening, claims = portClaims(config.portsDir, (line) => log("ports", line));
   const net = (input, init = {}) => fetch(input, { ...init, signal: init.signal ? AbortSignal.any([shutdown.signal, init.signal]) : shutdown.signal });
   let token = "";
   let supersede;
@@ -4609,7 +4681,7 @@ async function run3(config, bots = realBots) {
     if (!res.ok) throw new Error(`${init?.method ?? "GET"} ${path}: HTTP ${res.status} ${(await res.text()).slice(0, 200)}`);
     return res.json();
   };
-  const claim = await hub("/api/local-runner/claim", { method: "POST", body: JSON.stringify({ device: config.device, deviceName: config.deviceName, takeOver: config.takeOver, attachments: true }) });
+  const claim = await hub("/api/local-runner/claim", { method: "POST", body: JSON.stringify({ device: config.device, deviceName: config.deviceName, takeOver: config.takeOver, explicit: true, takeOverFrom: config.takeOverFrom, attachments: true }) });
   if (claim.status === 409) throw new HeldElsewhere((await claim.json()).holder);
   if (!claim.ok) throw new Error(`claim: HTTP ${claim.status} ${(await claim.text()).slice(0, 200)}`);
   const claimed = await claim.json();
@@ -4706,13 +4778,16 @@ async function run3(config, bots = realBots) {
     }
     await tell(null);
     const wanted = work.bots;
-    const taken = new Set([...running.values()].map((a) => a.port));
-    const fresh = wanted.filter((name) => !running.has(name)).map((name) => {
-      let port = config.portBase + work.bots.indexOf(name);
-      while (taken.has(port)) port++;
-      taken.add(port);
-      return { name, port };
-    });
+    const owned = adopted ? /* @__PURE__ */ new Map() : await supervisor.owned();
+    const fresh = await pickPorts(
+      wanted.filter((n) => !running.has(n)),
+      work.bots,
+      new Set([...running.values()].map((a) => a.port)),
+      config.portBase,
+      owned,
+      portBusy,
+      claims
+    );
     shutdown.signal.throwIfAborted();
     for (const agent of fresh) running.set(agent.name, agent);
     if (!adopted) {
@@ -4721,7 +4796,7 @@ async function run3(config, bots = realBots) {
     } else for (const agent of fresh) supervisor.start(agent);
     for (const [name, agent] of running) if (!wanted.includes(name)) {
       running.delete(name);
-      void supervisor.stop(agent);
+      void supervisor.stop(agent).then(() => claims.release(agent.port));
     }
     report2();
     await Promise.all(fresh.map(async (agent) => {
@@ -4975,6 +5050,7 @@ async function run3(config, bots = realBots) {
     model.stop();
     clearTimeout(retry);
     await supervisor.stopAll();
+    await claims.releaseAll();
     await Promise.allSettled([feed, syncing]);
   };
   void superseded.then(() => stopRunner(false));
@@ -4989,24 +5065,30 @@ async function run3(config, bots = realBots) {
 var need = (name) => process.env[name] || (() => {
   throw new Error(`${name} is required`);
 })();
-var configured = process.env.HUB_RUNNER_DIR || join16(homedir4(), ".hub", "local");
-mkdirSync7(configured, { recursive: true });
+var configured = process.env.HUB_RUNNER_DIR || join17(homedir4(), ".hub", "local");
+mkdirSync8(configured, { recursive: true });
 var dir = realpathSync4(configured);
-mkdirSync7(join16(dir, "status"), { recursive: true });
-statusFile(join16(dir, "status", `${process.pid}.json`), { pid: process.pid, state: "starting", build: process.env.HUB_RUNNER_BUILD ?? "" });
+mkdirSync8(join17(dir, "status"), { recursive: true });
+statusFile(join17(dir, "status", `${process.pid}.json`), { pid: process.pid, state: "starting", build: process.env.HUB_RUNNER_BUILD ?? "" });
 process.on("exit", exitStatus);
 var drainAsked = new Promise((resolve6) => process.on("SIGUSR2", () => resolve6()));
-var release = await import_proper_lockfile.default.lock(dir, { lockfilePath: join16(dir, "runner.lock"), stale: 1e4, realpath: false }).catch((error) => {
+var release = await import_proper_lockfile2.default.lock(dir, { lockfilePath: join17(dir, "runner.lock"), stale: 1e4, realpath: false }).catch((error) => {
   if (error.code !== "ELOCKED") throw error;
   status("busy-here");
   process.exit(5);
 });
-var deviceFile = join16(dir, "device-id");
+var deviceFile = join17(dir, "device-id");
 if (!existsSync11(deviceFile)) writeFileSync7(deviceFile, randomUUID());
-var device = readFileSync10(deviceFile, "utf8").trim();
-var credentials = process.env.HUB_URL || process.env.HUB_SESSION ? { hubUrl: need("HUB_URL"), session: need("HUB_SESSION"), bypass: process.env.HUB_PROTECTION_BYPASS || void 0 } : JSON.parse(readFileSync10(join16(dir, "credentials.json"), "utf8"));
+var device = readFileSync11(deviceFile, "utf8").trim();
+var credentials = process.env.HUB_URL || process.env.HUB_SESSION ? { hubUrl: need("HUB_URL"), session: need("HUB_SESSION"), bypass: process.env.HUB_PROTECTION_BYPASS || void 0 } : JSON.parse(readFileSync11(join17(dir, "credentials.json"), "utf8"));
 if ("workspace" in credentials && credentials.workspace) status("starting", { workspace: credentials.workspace });
 var deviceName = process.env.HUB_DEVICE_NAME || device;
+var takeOverFile = join17(dir, "take-over");
+var takeOverFromFile = join17(dir, "take-over-from");
+var takeOver = existsSync11(takeOverFile);
+var takeOverFrom = existsSync11(takeOverFromFile) ? readFileSync11(takeOverFromFile, "utf8").trim() || void 0 : void 0;
+rmSync5(takeOverFile, { force: true });
+rmSync5(takeOverFromFile, { force: true });
 var started = await run3({
   hubUrl: credentials.hubUrl.replace(/\/$/, ""),
   session: credentials.session,
@@ -5016,9 +5098,12 @@ var started = await run3({
   pnpm: need("HUB_PNPM"),
   // Two runners on one machine (the e2e) need their own ports; two Macs never share any.
   portBase: Number(process.env.HUB_RUNNER_PORT_BASE) || 5400,
+  // Shared by every workspace's runner on this Mac, whatever its dir: each claims its bots' ports there.
+  portsDir: process.env.HUB_RUNNER_PORTS_DIR || join17(homedir4(), ".hub", "ports"),
   device,
   deviceName,
-  takeOver: process.env.HUB_RUNNER_TAKE_OVER === "1"
+  takeOver,
+  takeOverFrom
 }).catch(async (error) => {
   if (!(error instanceof HeldElsewhere)) throw error;
   status("held-elsewhere", { holder: error.holder });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "hub-runner",
-  "version": "0.2.6",
+  "version": "0.2.7",
   "license": "Apache-2.0",
   "description": "Runs a hub workspace's Mac bots on this Mac under launchd, with no app needed.",
   "type": "module",