feat(server-s3): a member's tmux follows tmux upgrades — tmux-member + Users → Members' tmux
A member's tmux server keeps the version it started with until their last pane closes, but the dashboard asked it with TMUX_BIN, the newest tmux: a 3.8 client on a 3.6 server spun it at 100% CPU (three members needed a manual kill on 2026-10-15). - /opt/agenthub/libexec/tmux-member (install-system, in sudoers): runs a member's tmux call with the binary of THEIR running server (/proc/<pid>/exe, found from the socket in /proc/net/unix), else TMUX_BIN. It runs as the member and execs only a process they own. sudoArgv uses it once installed (linux-accounts readConf), TMUX_BIN until then. - Users → Members' tmux (GET /api/users/tmux): each member server's version beside the installed one, sessions, agents working. An admin restarts one (POST /api/users/:id/tmux-restart) or every idle outdated one (POST /api/users/tmux-restart-idle), with their password again, audited per member; a server that will not stop answers 409 ETMUXSTUCK naming its pid for root. - server-s3-system.test.js: `delete` is a provision action since Delete user, so the "unknown action" case now uses `purge`. Operators: run `sudo agenthub server install-system` once. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -46,6 +46,13 @@ function serverConf() {
|
||||
return confCache.value;
|
||||
}
|
||||
|
||||
let TMUX_MEMBER = "/opt/agenthub/libexec/tmux-member";
|
||||
/** Test seam: where tmux-member is looked for. */
|
||||
function __setTmuxMember(p) {
|
||||
TMUX_MEMBER = p;
|
||||
confCache = { at: 0, path: null, value: null };
|
||||
}
|
||||
|
||||
function readConf(file) {
|
||||
let text;
|
||||
try {
|
||||
@@ -56,16 +63,28 @@ function readConf(file) {
|
||||
const get = (k) => new RegExp(`^${k}=(.*)$`, "m").exec(text)?.[1]?.trim() || "";
|
||||
const usersRoot = get("USERS_ROOT");
|
||||
if (!path.isAbsolute(usersRoot)) return null;
|
||||
// A member's tmux runs through tmux-member once install-system put it there
|
||||
// (and in sudoers): it picks their RUNNING server's own binary, so a tmux
|
||||
// upgrade never mixes a new client with an old server. Before that, TMUX_BIN.
|
||||
let tmuxBin = get("TMUX_BIN");
|
||||
try {
|
||||
fs.accessSync(TMUX_MEMBER, fs.constants.X_OK);
|
||||
if (path.isAbsolute(tmuxBin)) tmuxBin = TMUX_MEMBER;
|
||||
} catch {
|
||||
/* not installed yet */
|
||||
}
|
||||
return {
|
||||
hostUser: get("HOST_USER"),
|
||||
hostUid: Number(get("HOST_UID")),
|
||||
usersRoot,
|
||||
port: Number(get("PORT")) || null,
|
||||
// The installed tmux (a new member server starts on it).
|
||||
tmuxInstalled: path.isAbsolute(get("TMUX_BIN")) ? get("TMUX_BIN") : null,
|
||||
// The absolute paths sudoers lists (sudo would resolve a bare name on its
|
||||
// own secure_path, which may not be them).
|
||||
bins: Object.fromEntries(
|
||||
[
|
||||
["tmux", get("TMUX_BIN")],
|
||||
["tmux", tmuxBin],
|
||||
["git", get("GIT_BIN")],
|
||||
["gh", get("GH_BIN")],
|
||||
].filter(([, v]) => path.isAbsolute(v))
|
||||
@@ -154,6 +173,7 @@ function __resetConfCache() {
|
||||
|
||||
module.exports = {
|
||||
__resetConfCache,
|
||||
__setTmuxMember,
|
||||
confPath,
|
||||
LINUX_NAME_RE,
|
||||
serverConf,
|
||||
|
||||
@@ -0,0 +1,103 @@
|
||||
/**
|
||||
* @file member-tmux.js
|
||||
* @description Server mode, one Linux user per account (S3): each member's
|
||||
* own tmux server as the Users page shows it — the version it runs next to
|
||||
* the installed one (`TMUX_BIN`), how many sessions it holds and whether an
|
||||
* agent of theirs is working — and an admin's restart of it. A tmux server
|
||||
* keeps the version it started with until its last session closes, so after
|
||||
* a tmux upgrade an old server lingers; `tmux-member` (install-system) keeps
|
||||
* every call on it matched to its own binary, and this is how an admin moves
|
||||
* a member onto the new version on purpose. A restart ends every pane of
|
||||
* theirs (their Claude sessions resume from the workspace).
|
||||
* @author Nguyễn Ngọc Trí Vĩ <vinnt@smartgift.vn>
|
||||
*/
|
||||
|
||||
const { execFileSync } = require("node:child_process");
|
||||
|
||||
const { db } = require("../db");
|
||||
const linux = require("./linux-accounts");
|
||||
const tmux = require("./tmux");
|
||||
|
||||
const err = (code, message) => Object.assign(new Error(message), { code });
|
||||
|
||||
// The installed tmux's version (`3.8`), re-read at most once a minute.
|
||||
let installed = { at: 0, bin: null, version: null };
|
||||
function installedVersion() {
|
||||
const bin = linux.serverConf()?.tmuxInstalled || null;
|
||||
if (installed.bin === bin && Date.now() - installed.at < 60_000) return installed.version;
|
||||
let version = null;
|
||||
if (bin) {
|
||||
try {
|
||||
version =
|
||||
String(execFileSync(bin, ["-V"], { encoding: "utf8", timeout: 3000, stdio: "pipe" }))
|
||||
.trim()
|
||||
.replace(/^tmux\s+/, "") || null;
|
||||
} catch {
|
||||
/* missing or broken: unknown */
|
||||
}
|
||||
}
|
||||
installed = { at: Date.now(), bin, version };
|
||||
return version;
|
||||
}
|
||||
|
||||
/** How many of `userId`'s workspaces have an agent working right now. */
|
||||
function busyCount(userId) {
|
||||
return db
|
||||
.prepare("SELECT COUNT(*) AS n FROM lanes WHERE owner_id = ? AND status = 'running'")
|
||||
.get(userId).n;
|
||||
}
|
||||
|
||||
/** One user's row; null for an account with no Linux side. */
|
||||
function rowFor(user, installedV) {
|
||||
const acct = linux.accountFor(user.username);
|
||||
if (!acct) return null;
|
||||
const info = tmux.memberServerInfo(acct);
|
||||
return {
|
||||
id: user.id,
|
||||
username: user.username,
|
||||
...info,
|
||||
busy: busyCount(user.id),
|
||||
outdated: !!(info.running && info.version && installedV && info.version !== installedV),
|
||||
};
|
||||
}
|
||||
|
||||
/** `{available, installed, matched, users: [...]}` for the Users page. */
|
||||
function overview(users) {
|
||||
if (!linux.s3Enabled()) return { available: false, users: [] };
|
||||
const conf = linux.serverConf();
|
||||
const v = installedVersion();
|
||||
return {
|
||||
available: true,
|
||||
installed: v,
|
||||
// Whether calls on an older server already use its own binary.
|
||||
matched: conf?.bins?.tmux !== conf?.tmuxInstalled,
|
||||
users: users.map((u) => rowFor(u, v)).filter(Boolean),
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Restart `user`'s tmux server. `{restarted: false}` when none ran; throws
|
||||
* ETMUXSTUCK (with the pid) when it would not stop — root has to kill it.
|
||||
*/
|
||||
function restart(user) {
|
||||
const acct = linux.accountFor(user.username);
|
||||
if (!acct) throw err("ENOS3", "this account has no Linux user (S3 is not installed)");
|
||||
const before = tmux.memberServerInfo(acct);
|
||||
if (!before.running) return { restarted: false, before };
|
||||
try {
|
||||
tmux.killMemberServer(acct);
|
||||
} catch {
|
||||
/* judged by what is left below */
|
||||
}
|
||||
const after = tmux.memberServerInfo(acct);
|
||||
if (after.running) {
|
||||
const pid = before.pid || after.pid;
|
||||
throw err(
|
||||
"ETMUXSTUCK",
|
||||
`${user.username}'s tmux server did not stop — run \`sudo kill -9 ${pid || "<pid>"}\` on the server`
|
||||
);
|
||||
}
|
||||
return { restarted: true, before };
|
||||
}
|
||||
|
||||
module.exports = { overview, restart, installedVersion };
|
||||
@@ -720,6 +720,53 @@ function killSession(name) {
|
||||
noteSession(name, accountOfQuiet(name), false);
|
||||
}
|
||||
|
||||
/**
|
||||
* A member's tmux server as the Users page shows it: `{running: false}`, or
|
||||
* `{running: true, version, pid, startedAt, sessions}`, or `{stuck: true}`
|
||||
* when it did not answer (hung, or backed off). Asked with `tmux-member`, i.e.
|
||||
* with that server's own binary.
|
||||
*/
|
||||
function memberServerInfo(acct) {
|
||||
try {
|
||||
const out = String(
|
||||
memberCall(acct, [
|
||||
"display-message",
|
||||
"-p",
|
||||
"#{version}\t#{pid}\t#{start_time}\t#{server_sessions}",
|
||||
])
|
||||
).trim();
|
||||
const [version, pid, start, sessions] = out.split("\t");
|
||||
return {
|
||||
running: true,
|
||||
version: version || null,
|
||||
pid: /^\d+$/.test(pid) ? Number(pid) : null,
|
||||
startedAt: /^\d+$/.test(start) ? Number(start) * 1000 : null,
|
||||
sessions: /^\d+$/.test(sessions) ? Number(sessions) : null,
|
||||
};
|
||||
} catch (err) {
|
||||
if (err.code === "ETMUXHUNG" || err.code === "ETIMEDOUT" || err.code === "ENOBUFS") {
|
||||
return { running: true, stuck: true };
|
||||
}
|
||||
// "no server running" / "error connecting": nothing to report.
|
||||
return { running: false };
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Stop a member's tmux server — every pane of theirs ends (an admin's Users
|
||||
* page action, so the next start runs the installed tmux). Throws when the
|
||||
* server did not go: a stuck one may need root (`sudo kill <pid>`).
|
||||
*/
|
||||
function killMemberServer(acct) {
|
||||
unreachable.delete(acct.login); // asked deliberately, past any back-off
|
||||
try {
|
||||
memberCall(acct, ["kill-server"]);
|
||||
} finally {
|
||||
listGen++;
|
||||
listed.delete(acct.login);
|
||||
}
|
||||
}
|
||||
|
||||
/** Session names starting with `prefix`. Empty array if tmux has no server running at all. */
|
||||
function listSessions(prefix) {
|
||||
const names = [];
|
||||
@@ -867,6 +914,8 @@ module.exports = {
|
||||
upgradeAdvice,
|
||||
serverPid,
|
||||
memberAttach,
|
||||
memberServerInfo,
|
||||
killMemberServer,
|
||||
__setExecImpl,
|
||||
__setAccountImpl,
|
||||
__settle,
|
||||
|
||||
Reference in New Issue
Block a user