Merge branch 's3-async-tmux' into s3-followups

docs/SERVER-MODE.md

@@ -567,7 +567,13 @@ a pane really is. Each case below is refused or bounded, with a test in
 - **A member's stopped or flooding tmux server.** Every call on it has a 3 s
   limit and a 1 MiB answer cap; one that hit either is not asked again for a
   minute. Meanwhile only that member's panes count as unknown (kept, not
-  ended); everyone else's are still checked. Every other synchronous call
+  ended); everyone else's are still checked. Listing a member's panes (every
+  poll: the pane list, the workspace list, the 10 s heal beat, the liveness
+  reaper) never waits for them: it runs in the background and the poll reads
+  its last answer, so a stopped server stalls nobody. Until a first answer,
+  or once the last is 15 s old, their panes count as unknown too. A call a
+  click makes on a member's pane (start, type, stop) still waits, up to the
+  3 s. Every other synchronous call
   that runs as a member (git in their folder, `write-as`) stops after 5 s.
 - **A member's huge or sparse files.** Transcripts and journals are read line
   by line with a 64 MiB cap per line (a 512 MB line used to crash the server);

docs/plans/server-mode/S3-PLAN.md

@@ -273,7 +273,7 @@ Review ran 2026-10-08 (four lenses: root/sudo, cross-member, secrets, local-mode
   - F34 a developer's extra workspace folder is checked against their own folder (`scope.folderAllowed`, real path) before anything probes it, and again on the path it records (`403 EFOLDER`);
   - F35 the Codex pool's quota reads a run's rollout only from the `CODEX_HOME` of the account its workspace runs as, never one dated in the future. Accepted ceiling (documented): a member can still forge their own session's reading.
   - F36 (critical) `GET /api/cc-config/file` never serves the data dir (`isSecretConfigPath`): an analyst read an admin gateway pane's live pane token from `proxy-settings/`, plus `vapid-keys.json` and the database;
-  - F37 every tmux call on a member's socket is bounded (`memberCall`: 3 s, SIGKILL, 1 MiB) with a 60 s back-off — a SIGSTOPped member server froze the event loop for good;
+  - F37 every tmux call on a member's socket is bounded (`memberCall`: 3 s, SIGKILL, 1 MiB) with a 60 s back-off — a SIGSTOPped member server froze the event loop for good. Member LISTING is async (`refreshList`: `execFile`, same limits) into the `listed` cache the sync readers use (stale-while-revalidate, 15 s max age, cold = unknown; a session this process makes or kills is written into the answer at once), so a frozen member no longer costs a 3 s block per back-off on the poll path; a click's call (new-session, send-keys, kill, has-session) is still a sync ≤ 3 s;
   - F38 a failing member server makes only that member's panes unknown (`liveSessions().has` true for them), never null for everyone; member list errors are cached;
   - F39 `execIn` and `write-as` (every sync sudo into a member) stop after 5 s (`MEMBER_EXEC`) — a FIFO `include.path` or a FIFO write target hung git and write-as;
   - F40 transcripts and journals are read with `readLines` (chunked, a line over 64 MiB is skipped, never built) — readline crashed the server on a one-line 512 MB sparse file (import-history, sessions, codex-usage);

server/__tests__/server-s3-panes.test.js

@@ -134,12 +134,24 @@ describe("a member's panes on the member's own tmux server", () => {
     assert.ok(calls.every((a) => a[0] !== "-S"));
   });
 
-  it("lists sessions across every server, believing a member's only about their own", () => {
+  it("lists sessions across every server, believing a member's only about their own", async () => {
     const orig = runAs.membersWithTmux;
     runAs.membersWithTmux = () => [ACCT];
     try {
+      // A member's list is async: until it lands their panes are UNKNOWN.
+      const cold = tmux.liveSessions();
+      assert.deepEqual([...cold].sort(), [HOST, "other"].sort());
+      assert.equal(cold.has(MEMBER), true, "unknown, kept");
+      assert.equal(cold.has("agenthub-lane-9"), false);
+      assert.deepEqual(tmux.listSessions("agenthub-lane-"), [HOST]);
+      await tmux.__settle();
       assert.deepEqual([...tmux.liveSessions()].sort(), [MEMBER, HOST, "other"].sort());
       assert.deepEqual(tmux.listSessions("agenthub-lane-").sort(), [HOST, MEMBER].sort());
+      // A session this process ends or makes is in the next answer at once.
+      tmux.killSession(MEMBER);
+      assert.deepEqual(tmux.listSessions("agenthub-lane-"), [HOST]);
+      tmux.newSession({ name: MEMBER, cwd: HOME, argv: ["bash"], env: {} });
+      assert.deepEqual(tmux.listSessions("agenthub-lane-").sort(), [HOST, MEMBER].sort());
     } finally {
       runAs.membersWithTmux = orig;
     }

server/__tests__/server-s3-review5.test.js

@@ -237,25 +237,46 @@ describe("a member's stopped tmux server or planted FIFO", () => {
     // sudo that never returns: a tmux client talking to a SIGSTOPped server.
     const bin = path.join(TMP, "fakebin");
     fs.mkdirSync(bin, { recursive: true });
-    fs.writeFileSync(path.join(bin, "sudo"), "#!/bin/sh\nexec sleep 60\n", { mode: 0o755 });
+    const spawned = path.join(TMP, "sudo-spawns");
+    fs.rmSync(spawned, { force: true });
+    fs.writeFileSync(path.join(bin, "sudo"), `#!/bin/sh\necho x >> '${spawned}'\nexec sleep 60\n`, {
+      mode: 0o755,
+    });
     try {
+      // Polls like the heal beat / GET /api/lanes for 4 s — past the 3 s
+      // timeout — while timing every poll and the event loop's worst stall.
       const child = inChild(
         `const runAs = require(${JSON.stringify(`${LIB}/run-as`)});
          const tmux = require(${JSON.stringify(`${LIB}/tmux`)});
+         const { monitorEventLoopDelay } = require("node:perf_hooks");
          const proj = ${JSON.stringify(path.join(MIA, "proj"))};
          const time = (f) => { const t = Date.now(); try { f(); return { ms: Date.now() - t }; }
            catch (e) { return { code: e.code, ms: Date.now() - t }; } };
+         (async () => {
+           runAs.accountForSession("agenthub-lane-${lane}"); // opens the DB: not what is timed
+           const loop = monitorEventLoopDelay({ resolution: 10 });
+           loop.enable();
            let live;
            const first = time(() => (live = tmux.liveSessions()));
-         const again = time(() => tmux.liveSessions());
+           let slowest = 0;
+           const beat = setInterval(() => {
+             slowest = Math.max(slowest, time(() => tmux.liveSessions()).ms);
+           }, 50);
+           await new Promise((r) => setTimeout(r, 4000));
+           clearInterval(beat);
+           loop.disable();
+           const after = tmux.liveSessions();
            process.stdout.write(JSON.stringify({
              isSet: live instanceof Set,
              keeps: live.has("agenthub-lane-${lane}"),
              other: live.has("agenthub-lane-999999"),
-           first, again,
+             keepsAfter: after.has("agenthub-lane-${lane}"),
+             first, slowest, stallMs: loop.max / 1e6,
+             sudos: require("node:fs").readFileSync(${JSON.stringify(spawned)}, "utf8").trim().split("\\n").length,
              git: time(() => runAs.execIn(proj, "git", ["status"])),
              write: time(() => runAs.writeAs(runAs.accountForPath(proj), proj + "/f.txt", "", "x")),
-         }));`,
+           }));
+         })();`,
         { PATH: `${bin}:${process.env.PATH}` }
       );
       assert.equal(child.signal, null, "nothing hung");
@@ -265,8 +286,14 @@ describe("a member's stopped tmux server or planted FIFO", () => {
       assert.equal(r.isSet, true, "the host's answer still counts");
       assert.equal(r.keeps, true);
       assert.equal(r.other, false);
-      assert.ok(r.first.ms < 10_000, `first list: ${r.first.ms} ms`);
-      assert.ok(r.again.ms < 500, `backed off: ${r.again.ms} ms`);
+      assert.equal(r.keepsAfter, true, "still unknown once the hung list was killed");
+      // Listing a member never waits for them: no poll, no event-loop turn
+      // takes anything like the 3 s a hung sudo is allowed.
+      assert.ok(r.first.ms < 100, `first list: ${r.first.ms} ms`);
+      assert.ok(r.slowest < 100, `slowest poll: ${r.slowest} ms`);
+      assert.ok(r.stallMs < 500, `event loop stalled ${r.stallMs} ms`);
+      // ONE sudo, killed at the timeout, then backed off — not one per poll.
+      assert.equal(r.sudos, 1);
       assert.equal(r.git.code, "ETIMEDOUT");
       assert.equal(r.write.code, "ETIMEDOUT");
     } finally {
@@ -505,9 +532,12 @@ describe("who a pane runs as", () => {
     assert.throws(() => auth.assignWorkspace(lane, "ana"), { code: "EOWNERFOLDER" });
     // One that got there anyway (assigned before this rule, or a demoted admin's).
     const theirs = addLane(path.join(HOST, "x"), "mia");
-    assert.throws(() => pty.spawnRun({ laneId: theirs, cwd: path.join(HOST, "x"), tool: "claude" }), {
+    assert.throws(
+      () => pty.spawnRun({ laneId: theirs, cwd: path.join(HOST, "x"), tool: "claude" }),
+      {
         code: "EOWNERFOLDER",
-    });
+      }
+    );
     assert.throws(() => pty.spawnShell({ laneId: theirs, cwd: path.join(HOST, "x") }), {
       code: "EOWNERFOLDER",
     });

server/lib/tmux.js

@@ -4,8 +4,9 @@
  * feature. Every dashboard-managed session is named `agenthub-lane-<id>` (see
  * `pty-run.js`) so a real terminal can attach to the exact same session with
  * `tmux attach -t agenthub-lane-<id>` (or `agenthub lanes shell`). Never builds a
- * shell string — every call is `execFileSync("tmux", [...argv])` with an
- * explicit argument array (matches this repo's rule for git in worktree.js).
+ * shell string — every call is `execFileSync("tmux", [...argv])` (a member's
+ * listing: `execFile`) with an explicit argument array (matches this repo's
+ * rule for git in worktree.js).
  * No call composes a shell command line at all, `sendCommand` included: it
  * takes over an idle pane with `respawn-pane -- <argv>` rather than typing a
  * quoted line into that pane's shell, which keeps an API key passed in `env`
@@ -21,13 +22,14 @@
  * that socket's folder, and a host client on it could be handed a fake server
  * or replaced by a member's command (`detach-client -E`). sudo logs its
  * command line, so a session's environment and argv travel in a launch file
- * (`run-as.launchArgv`) and a paste on stdin. Listing covers every server.
+ * (`run-as.launchArgv`) and a paste on stdin. Listing covers every server;
+ * a member's runs in the background and a poll reads its last answer.
  * With no member (local mode, S3 not installed, the host's own workspaces)
  * nothing changes.
  * @author Nguyễn Ngọc Trí Vĩ <vinnt@smartgift.vn>
  */
 
-const { execFileSync } = require("node:child_process");
+const { execFile, execFileSync } = require("node:child_process");
 
 // Test seam: swap the exec implementation so unit tests never invoke a real
 // tmux binary. Mirrors run-spawner.js's __injectChildForTest/__reset style.
@@ -67,39 +69,77 @@ let scopeImpl = realScopeImpl;
 // stdin). Seams, so tests never sudo.
 let accountImpl = (name) => require("./run-as").accountForSession(name);
 // A member's server is theirs: they can stop it (SIGSTOP) or fill it with
-// sessions, and every call here blocks the event loop. So each call on it is
-// bounded (time, output), and a server that timed out or flooded us is not
-// asked again for MEMBER_BACKOFF_MS — its panes read as UNKNOWN meanwhile.
-// ponytail: a frozen member still costs one MEMBER_TIMEOUT_MS block per
-// backoff; move member calls to async execFile if that ever shows.
+// sessions. So each call on it is bounded (time, output), and a server that
+// timed out or flooded us is not asked again for MEMBER_BACKOFF_MS — its
+// panes read as UNKNOWN meanwhile. LISTING (every poll: GET /api/run and
+// /api/lanes, the heal beat, the liveness reaper) never waits for a member:
+// it runs async (`memberListImpl`) into `listed`, which the poll reads. A
+// call a user's click makes (new-session, send-keys, kill, has-session)
+// stays synchronous.
+// ponytail: a click on a frozen member's pane (and a pane-token / gateway
+// has-session check) still blocks up to MEMBER_TIMEOUT_MS once per backoff;
+// make those async too if that ever shows.
 const MEMBER_TIMEOUT_MS = 3000;
 const MEMBER_BACKOFF_MS = 60_000;
-const MEMBER_OPTS = { ...EXEC_OPTS, timeout: MEMBER_TIMEOUT_MS, killSignal: "SIGKILL", maxBuffer: 1 << 20 };
-const realMemberImpl = (acct, args, input) => {
+const MEMBER_OPTS = {
+  ...EXEC_OPTS,
+  timeout: MEMBER_TIMEOUT_MS,
+  killSignal: "SIGKILL",
+  maxBuffer: 1 << 20,
+};
+const { stdio: _piped, ...MEMBER_LIST_OPTS } = MEMBER_OPTS; // execFile always pipes
+const sudoTmux = (acct, args) => {
   const runAs = require("./run-as");
-  const [cmd, argv] = runAs.sudoArgv(acct, "tmux", ["-S", runAs.tmuxSocket(acct), ...args]);
+  return runAs.sudoArgv(acct, "tmux", ["-S", runAs.tmuxSocket(acct), ...args]);
+};
+const realMemberImpl = (acct, args, input) => {
+  const [cmd, argv] = sudoTmux(acct, args);
   return execFileSync(cmd, argv, input == null ? MEMBER_OPTS : { ...MEMBER_OPTS, input });
 };
+// The same, async, with execFileSync's error shape: `status` = exit code,
+// ETIMEDOUT / ENOBUFS for a kill at the timeout / a flood.
+const realMemberListImpl = (acct, args) =>
+  new Promise((resolve, reject) => {
+    const [cmd, argv] = sudoTmux(acct, args);
+    execFile(cmd, argv, MEMBER_LIST_OPTS, (err, out) => {
+      if (!err) return resolve(out);
+      if (typeof err.code === "number") err.status = err.code;
+      if (err.killed) err.code = "ETIMEDOUT";
+      if (err.code === "ERR_CHILD_PROCESS_STDIO_MAXBUFFER") err.code = "ENOBUFS";
+      reject(err);
+    });
+  });
 let memberImpl = realMemberImpl;
+let memberListImpl = realMemberListImpl;
 let unreachable = new Map(); // member login → backoff end
-function memberCall(acct, args, input) {
+function reachable(acct) {
   if ((unreachable.get(acct.login) || 0) > Date.now()) {
     throw Object.assign(new Error(`${acct.login}: tmux server not answering`), {
       code: "ETMUXHUNG",
     });
   }
-  try {
-    return memberImpl(acct, args, input);
-  } catch (err) {
+}
+function noteHang(acct, err) {
   if (err.code === "ETIMEDOUT" || err.code === "ENOBUFS") {
     unreachable.set(acct.login, Date.now() + MEMBER_BACKOFF_MS);
   }
+}
+function memberCall(acct, args, input) {
+  reachable(acct);
+  try {
+    return memberImpl(acct, args, input);
+  } catch (err) {
+    noteHang(acct, err);
     throw err;
   }
 }
+/** Test seam. A sync `member` fake also answers the async listing. */
 function __setAccountImpl(fn, member) {
   accountImpl = fn;
-  if (member) memberImpl = member;
+  if (member) {
+    memberImpl = member;
+    memberListImpl = (acct, args) => new Promise((resolve) => resolve(member(acct, args)));
+  }
 }
 function accountOf(name) {
   try {
@@ -128,6 +168,7 @@ function __setExecImpl(fn, { exact = true } = {}) {
 
   exactTargets = exact;
   listed = new Map();
+  listGen++;
 }
 function __setScopeImpl(fn) {
   scopeImpl = fn;
@@ -137,8 +178,10 @@ function __reset() {
   scopeImpl = realScopeImpl;
   accountImpl = (name) => require("./run-as").accountForSession(name);
   memberImpl = realMemberImpl;
+  memberListImpl = realMemberListImpl;
   exactTargets = null;
   listed = new Map();
+  listGen++;
   unreachable = new Map();
 }
 
@@ -149,34 +192,90 @@ function __reset() {
 // own can read as live for up to a second. "No server" (exit 1) is an answer
 // too and is shared; any other failure is not.
 const LIST_TTL_MS = 1000;
+const LIST_ARGS = ["list-sessions", "-F", "#{session_name}"];
 let listed = new Map(); // member login ("" = ours) → { at, out } | { at, err }
+let listGen = 0; // bumped by a change this process makes: drops lists in flight
 function listOutput(acct = null) {
-  const key = acct ? acct.login : "";
-  const hit = listed.get(key);
+  if (acct) return memberListOutput(acct);
+  const hit = listed.get("");
   if (hit && Date.now() - hit.at < LIST_TTL_MS) {
     if (hit.err) throw hit.err;
     return hit.out;
   }
   try {
-    const list = ["list-sessions", "-F", "#{session_name}"];
-    let out = String(acct ? memberCall(acct, list) : execImpl(list));
+    const out = String(execImpl(LIST_ARGS));
+    listed.set("", { at: Date.now(), out });
+    return out;
+  } catch (err) {
+    if (err.status === 1 || err.code === "ENOENT") listed.set("", { at: Date.now(), err });
+    else listed.delete("");
+    throw err;
+  }
+}
+
+// A member's list is never waited for: a read answers from `listed` and,
+// once that is LIST_TTL_MS old, starts one async re-list. Until a first
+// answer — or once the last is MEMBER_LIST_MAX_AGE_MS old (nobody polled) —
+// or while their server fails, it throws: their panes are UNKNOWN. A session
+// this process creates or kills is written into the answer at once (`noteSession`).
+const MEMBER_LIST_MAX_AGE_MS = 15_000; // > the 10 s heal beat
+const refreshing = new Set(); // member logins with a list in flight
+function memberListOutput(acct) {
+  const hit = listed.get(acct.login);
+  const age = hit ? Date.now() - hit.at : Infinity;
+  if (age >= LIST_TTL_MS) refreshList(acct);
+  if (age >= MEMBER_LIST_MAX_AGE_MS) {
+    throw Object.assign(new Error(`${acct.login}: tmux list pending`), { code: "ETMUXPENDING" });
+  }
+  if (hit.err) throw hit.err;
+  return hit.out;
+}
+function refreshList(acct) {
+  const key = acct.login;
+  if (refreshing.has(key)) return;
+  refreshing.add(key);
+  const gen = listGen;
+  const hangs = unreachable; // a test's __reset starts a new one
+  const done = (entry) => {
+    refreshing.delete(key);
+    if (gen === listGen) listed.set(key, { at: Date.now(), ...entry });
+  };
+  new Promise((resolve) => {
+    reachable(acct);
+    resolve(memberListImpl(acct, LIST_ARGS));
+  }).then(
     // A member's server answers for their own workspaces only: a name it
     // lists that belongs to anyone else (a server the member runs to lie
     // about another pane) is not believed.
-    if (acct)
-      out = out
+    (out) =>
+      done({
+        out: String(out)
           .split("\n")
           .filter((n) => ownedBy(n.trim(), acct))
-        .join("\n");
-    listed.set(key, { at: Date.now(), out });
-    return out;
-  } catch (err) {
-    // A member's failure is shared too: a broken server of theirs is not
-    // asked again on every poll.
-    if (acct || err.status === 1 || err.code === "ENOENT") listed.set(key, { at: Date.now(), err });
-    else listed.delete(key);
-    throw err;
+          .join("\n"),
+      }),
+    (err) => {
+      if (hangs === unreachable) noteHang(acct, err);
+      // "No server" is an empty list; any other failure is shared too, so a
+      // broken server of theirs is not asked again on every poll.
+      done(err.status === 1 || err.code === "ENOENT" ? { out: "" } : { err });
     }
+  );
+}
+/** A session this process made or ended: ours re-lists, a member's answer is
+ *  corrected in place (an unknown one stays unknown). */
+function noteSession(name, acct, alive) {
+  listGen++;
+  listed.delete("");
+  const hit = acct && listed.get(acct.login);
+  if (!hit || hit.err) return;
+  const names = hit.out.split("\n").filter((n) => n.trim() && n.trim() !== name);
+  if (alive) names.push(name);
+  listed.set(acct.login, { at: hit.at, out: names.join("\n") });
+}
+/** Test seam: resolves once no member list is in flight. */
+async function __settle() {
+  while (refreshing.size) await new Promise((r) => setImmediate(r));
 }
 
 function ownedBy(name, acct) {
@@ -273,7 +372,7 @@ function envFlags(env) {
 function newSession({ name, cwd, argv, env }) {
   // No argv = tmux's own default shell (a Windows Terminal pane).
   const cmd = argv.length ? ["--", ...argv] : [];
-  listed = new Map();
+  noteSession(name, null, false);
   const args = ["new-session", "-d", "-s", name, "-c", cwd, ...envFlags(env), ...cmd];
   const acct = accountOf(name);
   if (acct) {
@@ -281,6 +380,7 @@ function newSession({ name, cwd, argv, env }) {
     // the launch file, never on sudo's logged command line.
     const launch = require("./run-as").launchArgv(acct, env, argv);
     memberCall(acct, ["new-session", "-d", "-s", name, "-c", cwd, "--", ...launch]);
+    noteSession(name, acct, true);
   } else if (scopeImpl && process.platform === "linux" && !serverPid()) {
     try {
       scopeImpl(args);
@@ -449,12 +549,12 @@ function pasteText(name, text) {
 
 /** Idempotent — a session that's already gone is not an error. */
 function killSession(name) {
-  listed = new Map();
   try {
     onServerOf(name, ["kill-session", "-t", session(name)]);
   } catch {
     /* already gone */
   }
+  noteSession(name, accountOfQuiet(name), false);
 }
 
 /** Session names starting with `prefix`. Empty array if tmux has no server running at all. */
@@ -567,5 +667,6 @@ module.exports = {
   memberAttach,
   __setExecImpl,
   __setAccountImpl,
+  __settle,
   __reset,
 };