lib: fix cluster handle leak

It is possible to cause a resource leak in SharedHandle if a worker is
added after all other workers have been removed. This commit fixes the
leak.

Fixes: nodejs#2510

lib/cluster.js

@@ -345,7 +345,10 @@ function masterInit() {
        * if it has disconnected, otherwise we might
        * still want to access it.
        */
-      if (!worker.isConnected()) removeWorker(worker);
+      if (!worker.isConnected()) {
+        removeHandlesForWorker(worker);
+        removeWorker(worker);
+      }
 
       worker.suicide = !!worker.suicide;
       worker.state = 'dead';

test/parallel/test-cluster-shared-worker-added-after-disconnect.js

@@ -0,0 +1,50 @@
+'use strict';
+const common = require('../common');
+const assert = require('assert');
+const net = require('net');
+const cluster = require('cluster');
+cluster.schedulingPolicy = cluster.SCHED_NONE;
+
+if (cluster.isMaster) {
+  var shotOnce = false;
+
+  function shoot() {
+    var c = net.connect(common.PORT, function() {
+      c.unref();
+
+      Object.keys(cluster.workers).forEach(function(id) {
+        cluster.workers[id].send('die');
+      });
+    });
+  }
+
+  function fork() {
+    var worker = cluster.fork();
+    worker.on('message', function() {
+      if (!shotOnce) {
+        shotOnce = true;
+        shoot();
+      }
+      fork();
+    });
+  }
+  fork();
+  return;
+}
+
+var server = net.createServer({}, function(c) {
+  c.end();
+});
+
+server.listen(common.PORT, function() {
+  process.send('listening');
+});
+
+process.on('message', function listener(msg) {
+  if (msg === 'die') {
+    server.close(function() {
+      process.exit();
+    });
+  }
+});
+