@volter/supercode-teams 0.2.98 → 0.2.100

package/bin/workspace.mjs

@@ -19,6 +19,9 @@ import { duration, parseArgs, positiveInteger, requireArgs, usage } from './work
 import { reserveArtifactDirectory, reserveProtected, writeProtected } from './workspace/files.mjs';
 import { acquireLifecycle, activeSocket, lifecycleHeld, requestStop } from './workspace/lifecycle.mjs';
 import { retireLinklessDaemon, watchCodeChange } from './workspace/upgrade-watch.mjs';
+
+/** How long a service-run connector's machine link may stay down before the connector restarts itself. */
+const LINK_LOST_MS = 120_000;
 import { volterDeviceSignIn } from './workspace/volter-device.mjs';
 import { loginSummary, signInToTeams } from './workspace/volter-login.mjs';
 import { readSharedSignIn, signOut } from '@volter/identity';
@@ -733,6 +736,15 @@ async function connectorCommand(verb, { flags, rest, output, client, context, te
     // A connected machine is a person's: each pane it launches opens a window on its screen (SUPERCODE_POPUPS=0: none).
     const machine = new TeamsMachine({ home: daemonHome, supercode: binary, popups: process.env.SUPERCODE_POPUPS !== '0', ...(flags.tmux_socket ? { tmuxSocket: String(flags.tmux_socket) } : {}), link: { server: context.origin, teamId: context.team_id, machineId: saved.machine.id, credential: saved.credential.token, incarnationId: connector.state.incarnation_id, configuration: () => connector.configuration() }, onTurn: (turn) => connector.noteTurn(turn), onTurnWithdrawn: (turn) => connector.withdrawTurn(turn) });
     machine.on('link', (event) => { if (event.event === 'replaced') process.stderr.write(`supercode teams: ${event.error}. Two connectors were running for this machine; stop one of them (\`supercode teams disconnect\` where it should not run).\n`); });
+    // The link's state for the watch after start: down until it is up, down again (since when) when it drops. A
+    // `replaced` link is left alone: another connector for this machine owns it, by design.
+    let linkDownSince = Date.now();
+    let linkReplaced = false;
+    machine.on('link', (event) => {
+      if (event.event === 'up') linkDownSince = null;
+      else if (event.event === 'dropped') linkDownSince ??= Date.now();
+      else if (event.event === 'replaced') linkReplaced = true;
+    });
     // The connector is the service's daemon and keeps serving; a daemon that took its socket path is the one to stop.
     machine.on('superseded', (path) => process.stderr.write(`supercode teams: another machine daemon replaced this connector's socket at ${path}; stop it (supercode teams status)\n`));
     const stopAll = async () => { await machine.stop().catch(() => {}); await connector.stop(); await core.close(); await largeCore.close(); await slowCore.close(); await liveCore.close(); await lifecycle.close(); state.update(current => current ? ({ ...current, process: undefined }) : null); };
@@ -757,9 +769,24 @@ async function connectorCommand(verb, { flags, rest, output, client, context, te
         process.stderr.write(`supercode teams: ${changed} changed under this connector (an upgrade); moving to the new code\n`);
         const cli = process.platform === 'win32' ? 'supercode' : process.env.SUPERCODE_BIN;
         spawn(cli, ['teams', 'connect', '--install', '--context', contextName, '--cwd', path], { shell: process.platform === 'win32', detached: true, stdio: 'ignore', windowsHide: true }).unref();
-        if (process.platform !== 'win32') setTimeout(() => { process.exitCode = 75; void stopConnector?.(); }, 60_000);
+        // An explicit exit: the command's own return would end the process with 0, which systemd's on-failure policy
+        // does not restart.
+        if (process.platform !== 'win32') setTimeout(async () => { await stopConnector?.(); process.exit(75); }, 60_000);
       });
     }
+    // Run by a service, a connector whose machine link stays down restarts itself, so a machine is never left
+    // unreachable with no remote way back (the Mac mini sat unlinked for hours while its connector lived on): a
+    // daemon that died, a start that half failed or a link its backoff never recovered all end in a fresh start.
+    // It exits non-zero, which every service manager here restarts.
+    if (saved.service) {
+      const linkWatch = setInterval(() => {
+        if (linkReplaced || linkDownSince === null || Date.now() - linkDownSince < LINK_LOST_MS) return;
+        clearInterval(linkWatch);
+        process.stderr.write(`supercode teams: this machine has had no link to its Teams server for ${Math.round((Date.now() - linkDownSince) / 1000)} s; restarting the connector\n`);
+        void (async () => { await stopConnector?.(); process.exit(75); })();
+      }, 15_000);
+      linkWatch.unref?.();
+    }
     await new Promise(resolveStop => { let stopped = false; stopConnector = async () => { if (stopped) return; stopped = true; await stopAll(); resolveStop(); }; for (const signal of ['SIGINT', 'SIGTERM', ...(process.platform === 'win32' ? ['SIGHUP', 'SIGBREAK'] : [])]) process.once(signal, stopConnector); });
     return;
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@volter/supercode-teams",
-  "version": "0.2.98",
+  "version": "0.2.100",
   "type": "module",
   "description": "A machine daemon serving this machine's doors, and a local or remote Teams server for shared session discovery, OpenTelemetry ingestion and access to enrolled machines.",
   "license": "MIT",