@mindexec/cli 0.2.140 → 0.2.142

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindexec/cli",
-  "version": "0.2.140",
+  "version": "0.2.142",
   "description": "MindExec local runtime and bridge CLI",
   "main": "server.js",
   "type": "module",

package/scripts/remote-agent-managed-smoke.mjs

@@ -124,6 +124,30 @@ function spawnBridge({ bridgePort, remoteHubPort, workspacePath, label }) {
     };
 }
 
+async function startHangingRemoteManager(port) {
+    const sockets = new Set();
+    const server = net.createServer(socket => {
+        sockets.add(socket);
+        socket.on('close', () => sockets.delete(socket));
+        socket.on('error', () => sockets.delete(socket));
+    });
+    server.unref();
+    await new Promise((resolve, reject) => {
+        server.once('error', reject);
+        server.listen(port, '127.0.0.1', resolve);
+    });
+
+    return {
+        endpoint: `127.0.0.1:${port}`,
+        async stop() {
+            for (const socket of sockets) {
+                socket.destroy();
+            }
+            await new Promise(resolve => server.close(resolve));
+        }
+    };
+}
+
 async function waitForBridge(bridge) {
     return await waitFor(async () => {
         const result = await fetchJson(`${bridge.baseUrl}/api/status`);
@@ -159,11 +183,13 @@ async function main() {
     const clientRemotePort = await findFreePort();
     const stalePort = await findFreePort();
     const managerEndpoint = `127.0.0.1:${hostRemotePort}`;
+    let hangingManager = null;
     const staleEndpoint = `127.0.0.1:${stalePort}`;
 
     let hostBridge = null;
     let clientBridge = null;
     try {
+        hangingManager = await startHangingRemoteManager(stalePort);
         hostBridge = spawnBridge({
             bridgePort: hostBridgePort,
             remoteHubPort: hostRemotePort,
@@ -180,11 +206,16 @@ async function main() {
         });
         await waitForBridge(clientBridge);
 
+        const firstConnectStartedAt = Date.now();
         const firstAgent = await connectManagedAgent(clientBridge, managerEndpoint, staleEndpoint);
+        const firstConnectMs = Date.now() - firstConnectStartedAt;
         assert.equal(firstAgent.running, true, JSON.stringify(firstAgent));
         assert.equal(firstAgent.ready, true, JSON.stringify(firstAgent));
         assert.equal(firstAgent.usingNpx, false, JSON.stringify(firstAgent));
         assert.match(String(firstAgent.launcher || ''), /mindexec-remote-fast/i);
+        assert.ok(
+            firstConnectMs < 4500,
+            `managed RemoteAgent must race slow endpoint candidates instead of waiting sequentially, got ${firstConnectMs}ms`);
 
         const connectedDevice = await waitFor(async () => {
             const result = await fetchJson(`${hostBridge.baseUrl}/api/remote/devices`);
@@ -218,6 +249,9 @@ async function main() {
 
         console.log('RemoteAgent managed supervisor smoke OK');
     } finally {
+        if (hangingManager) {
+            await hangingManager.stop();
+        }
         if (clientBridge) {
             await clientBridge.stop();
         }

package/scripts/remote-registry-follower-smoke.mjs

@@ -317,6 +317,20 @@ async function waitForManagedAgent(bridge, managerEndpoint, label, timeoutMs = 2
     }, timeoutMs, `${label} managed agent\n${bridge.details()}`);
 }
 
+async function waitForManagedAgentRestart(bridge, managerEndpoint, previousPid, label, timeoutMs = 20000) {
+    return await waitFor(async () => {
+        const result = await fetchJson(`${bridge.baseUrl}/api/remote/agent/status`);
+        const agent = result.payload;
+        return agent?.running === true
+            && agent?.ready === true
+            && String(agent.manager || '') === managerEndpoint
+            && Number(agent.pid || 0) > 0
+            && Number(agent.pid || 0) !== Number(previousPid || 0)
+            ? agent
+            : null;
+    }, timeoutMs, `${label} managed agent restart\n${bridge.details()}`);
+}
+
 async function waitForConnectedDevice(bridge, label) {
     return await waitFor(async () => {
         const result = await fetchJson(`${bridge.baseUrl}/api/remote/devices`);
@@ -324,6 +338,18 @@ async function waitForConnectedDevice(bridge, label) {
     }, 12000, `${label} connected device\n${bridge.details()}`);
 }
 
+function killProcess(pid) {
+    const value = Number(pid || 0);
+    assert.ok(value > 0, `expected positive pid, got ${pid}`);
+    try {
+        process.kill(value, 'SIGTERM');
+    } catch (error) {
+        if (error?.code !== 'ESRCH') {
+            throw error;
+        }
+    }
+}
+
 async function main() {
     const tempRoot = await mkdtemp(path.join(os.tmpdir(), 'mindexec-remote-registry-smoke-'));
     let fakeSupabase = null;
@@ -407,6 +433,14 @@ async function main() {
         assert.equal(secondAgent.leaseId, 'lease-b');
         await waitForConnectedDevice(hostB, 'host-b');
 
+        const secondPid = Number(secondAgent.pid || 0);
+        killProcess(secondPid);
+        const restartedAgent = await waitForManagedAgentRestart(client, hostBEndpoint, secondPid, 'host-b', 12000);
+        assert.equal(restartedAgent.usingNpx, false, JSON.stringify(restartedAgent));
+        assert.match(String(restartedAgent.launcher || ''), /mindexec-remote-fast/i);
+        assert.equal(restartedAgent.leaseId, 'lease-b');
+        await waitForConnectedDevice(hostB, 'host-b after restart');
+
         currentTarget = {
             ...currentTarget,
             active: false,

package/server.js

@@ -3132,7 +3132,7 @@ const REMOTE_AGENT_EARLY_EXIT_MS = 1200;
 const REMOTE_AGENT_READY_TIMEOUT_MS = 5000;
 const REMOTE_AGENT_DEFAULT_ENGINE = 'auto';
 const REMOTE_AGENT_SYNC_REPORT_LOG_REPEAT_MS = 30000;
-const REMOTE_AGENT_RACE_START_STAGGER_MS = 120;
+const REMOTE_AGENT_RACE_START_STAGGER_MS = Math.max(0, Number(process.env.MINDEXEC_REMOTE_AGENT_RACE_STAGGER_MS ?? 0) || 0);
 const REMOTE_AGENT_MAX_PARALLEL_CANDIDATES = 6;
 const REMOTE_AGENT_RECENT_MANAGER_CACHE_TTL_MS = 30 * 24 * 60 * 60 * 1000;
 const REMOTE_AGENT_RECENT_MANAGER_CACHE_LIMIT = 64;
@@ -3151,6 +3151,7 @@ let remoteAgentSyncReportState = null;
 let remoteAgentSyncReportLogKey = '';
 let remoteAgentSyncReportLogAt = 0;
 let remoteAgentConnectPromise = null;
+const remoteAgentIntentionalStopKeys = new Set();
 const remoteAgentRecentSuccessfulManagers = new Map();
 let remoteAgentRecentSuccessfulManagersLoaded = false;
 let remoteAgentRecentSuccessfulManagersSavePromise = null;
@@ -3656,6 +3657,36 @@ function isRemoteAgentRegistryOwned() {
         && /registry/i.test(String(remoteAgentState.source || ''));
 }
 
+function isRemoteAgentRegistryState(state) {
+    return /registry/i.test(String(state?.source || ''));
+}
+
+function suppressRemoteAgentAutoRetry(connectionKey) {
+    const key = safeRemoteAgentField(connectionKey, 128);
+    if (!key) {
+        return;
+    }
+
+    remoteAgentIntentionalStopKeys.add(key);
+    const timer = setTimeout(() => {
+        remoteAgentIntentionalStopKeys.delete(key);
+    }, 10000);
+    timer?.unref?.();
+}
+
+function maybeRetryRemoteAgentAfterExit(state, reason = 'agent-exited') {
+    if (!state
+        || !REMOTE_REGISTRY_FOLLOWER_ENABLED
+        || isShuttingDown
+        || !isRemoteAgentRegistryState(state)
+        || remoteAgentIntentionalStopKeys.has(String(state.connectionKey || ''))) {
+        return;
+    }
+
+    logWarn('remote', `managed RemoteAgent registry process ended; waking registry follower ${formatKeyValue('reason', reason)} ${formatKeyValue('manager', state.manager || '-')}`);
+    scheduleRemoteRegistryFollower(0, reason);
+}
+
 function isLocalRemoteHostTargetActive() {
     const status = remoteHub.getStatus({ includeSecrets: false });
     return status?.hostTargetActive === true
@@ -3912,6 +3943,7 @@ function resolveNpxCliLauncher() {
 
 async function stopRemoteAgentConnection(reason = 'stopped') {
     const previous = remoteAgentState;
+    suppressRemoteAgentAutoRetry(previous.connectionKey);
     if (previous.proc) {
         try {
             await terminateProcessTree(previous.proc);
@@ -4143,6 +4175,17 @@ function createRemoteAgentAttempt(options = {}) {
                 state.lastError = signal ? `signal:${signal}` : `exit:${code}`;
             }
             attempt.error = attempt.error || state.lastError || `exit:${code ?? 'unknown'}`;
+            if (remoteAgentState === state
+                && remoteAgentState.connectionKey === state.connectionKey) {
+                const serialized = serializeRemoteAgentState();
+                if (state.status === 'failed') {
+                    logWarn('remote', `managed RemoteAgent failed: ${formatRemoteAgentFailureSummary(state)}`);
+                } else {
+                    logEvent('remote', `managed RemoteAgent exited ${formatKeyValue('code', code ?? 0)}`, 'remote');
+                }
+                emitBridgeEvent(state.status === 'exited' ? 'RemoteAgentExited' : 'RemoteAgentFailed', serialized);
+                maybeRetryRemoteAgentAfterExit(state, state.status === 'exited' ? 'agent-exited' : 'agent-failed');
+            }
         });
     } catch (err) {
         attempt.failed = true;