fix: node orphan process detection and exit logix to fix mac node termination on sleep

Author: abose

package-lock.json

@@ -103,10 +103,6 @@ const rl = readline.createInterface({
 let serverPortResolve;
 const serverPortPromise = new Promise((resolve) => { serverPortResolve = resolve; });
 
-let orphanExitTimer = setTimeout(()=>{
-    process.exit(1);
-}, 60000);
-
 function _sendResponse(responseMessage, commandID) {
     savedConsoleLog(COMMAND_RESPONSE_PREFIX + JSON.stringify({
         message: responseMessage,
@@ -153,15 +149,16 @@ process.on('uncaughtException', (error) => {
     _sendError(error, "uncaughtException");
 });
 
-function resetOrphanExitTimer() {
-    const timeout = debugMode ? 60000 * 15 : 60000;
-    // in debug mode, we wait for more minutes to not exit node if phcode doesn't send heartbeats on break point debug
-    clearTimeout(orphanExitTimer);
-    orphanExitTimer = setTimeout(()=>{
-        console.error(`Node did not receive heartbeat for ${timeout/1000}S, existing as orphan node process detected.`);
-        process.exit(1);
-    }, timeout);
-}
+// In the child process
+process.on('disconnect', () => {
+    console.log('Parent process exited, exiting.');
+    process.exit(1);
+});
+
+rl.on('close', () => {
+    console.log('Parent input stream is closed');
+    process.exit(1);
+});
 
 function processCommand(line) {
     try{
@@ -175,9 +172,6 @@ function processCommand(line) {
                     process.exit(0);
                 });
             return;
-        case "heartBeat":
-            resetOrphanExitTimer();
-            return;
         case "ping": _sendResponse("pong", jsonCmd.commandID); return;
         case "setDebugMode":
             debugMode = jsonCmd.commandData;
@@ -187,7 +181,6 @@ function processCommand(line) {
             } else {
                 console.log = function () {}; // swallow logs
             }
-            resetOrphanExitTimer();
             _sendResponse("done", jsonCmd.commandID); return;
         case "getEndpoints":
             serverPortPromise.then(port =>{

src-node/index.js

@@ -105,6 +105,24 @@ const LARGE_DATA_THRESHOLD = 2*1024*1024; // 2MB
 let controlSocketMain = null,
     largeDataSocketMain = null;
 
+const SOCKET_HEALTH_CHECK_INTERVAL_MS = 10 * 1000; // Interval in milliseconds
+let healthCheckFailuresCount = 0; // Count of consecutive health check failures
+const RETRY_HEALTH_CHECK_MAX_COUNT = 2;
+setInterval(() => {
+    if (healthCheckFailuresCount === RETRY_HEALTH_CHECK_MAX_COUNT) {
+        console.error(`No socket connection for ${healthCheckFailuresCount * SOCKET_HEALTH_CHECK_INTERVAL_MS / 1000} seconds. Exiting due to orphan node process detection.`);
+        process.exit(1);
+    }
+
+    if (controlSocketMain || largeDataSocketMain) {
+        // Reset failure count if any socket is connected
+        healthCheckFailuresCount = 0;
+        return;
+    }
+    console.warn(`No active sockets detected, node will be terminated in ${(RETRY_HEALTH_CHECK_MAX_COUNT-healthCheckFailuresCount)*SOCKET_HEALTH_CHECK_INTERVAL_MS/1000} seconds`);
+    healthCheckFailuresCount++;
+}, SOCKET_HEALTH_CHECK_INTERVAL_MS);
+
 const MAX_PENDING_SEND_BUFFER = 10000;
 let pendingSendBuffer = [];
 

src-node/node-connector.js

@@ -178,6 +178,16 @@ function nodeLoader() {
         return commandID;
     }
 
+    function _sendInitCommand(socket, commandCode) {
+        currentCommandID++;
+        const commandID = currentCommandID;
+        socket.send(mergeMetadataAndArrayBuffer({
+            commandCode: commandCode,
+            commandID: commandID,
+            data: null
+        }, null));
+    }
+
     function _sendExec(nodeConnectorID, commandID, execHandlerFnName, dataObjectToSend = null, dataBuffer = null) {
         const command = {
             nodeConnectorID: nodeConnectorID,
@@ -521,9 +531,9 @@ function nodeLoader() {
                     firstConnectCB();
                 }
                 if(ws.isLargeDataWS){
-                    _sendCommand(WS_COMMAND.LARGE_DATA_SOCKET_ANNOUNCE);
+                    _sendInitCommand(ws, WS_COMMAND.LARGE_DATA_SOCKET_ANNOUNCE);
                 } else {
-                    _sendCommand(WS_COMMAND.CONTROL_SOCKET_ANNOUNCE);
+                    _sendInitCommand(ws, WS_COMMAND.CONTROL_SOCKET_ANNOUNCE);
                 }
                 _drainPendingSendBuffer();
             });
@@ -582,7 +592,6 @@ function nodeLoader() {
             TERMINATE: "terminate",
             PING: "ping",
             SET_DEBUG_MODE: "setDebugMode",
-            HEART_BEAT: "heartBeat",
             GET_ENDPOINTS: "getEndpoints"
         };
         const COMMAND_RESPONSE_PREFIX = 'phnodeResp_1!5$:'; // a string thats not likely to just start with in
@@ -709,11 +718,6 @@ function nodeLoader() {
                         window.PhNodeEngine._nodeLoadTime = Date.now() - nodeLoadstartTime;
                     });
                 execNode(NODE_COMMANDS.SET_DEBUG_MODE, window.debugMode);
-                setInterval(()=>{
-                    if(!window.isNodeTerminated) {
-                        execNode(NODE_COMMANDS.HEART_BEAT);
-                    }
-                }, 10000);
             });
     });
 }