[FIXED JENKINS-38539] Turn on SO_KEEPALIVE and provide CLI option to turn it off again

src/main/java/hudson/remoting/Engine.java

@@ -138,6 +138,13 @@ public void run() {
 
     private boolean noReconnect;
 
+    /**
+     * Determines whether the socket will have {@link Socket#setKeepAlive(boolean)} set or not.
+     *
+     * @since TODO
+     */
+    private boolean keepAlive = true;
+
     private JarCache jarCache = new FileSystemJarCache(new File(System.getProperty("user.home"),".jenkins/cache/jars"),true);
 
     public Engine(EngineListener listener, List<URL> hudsonUrls, String secretKey, String slaveName) {
@@ -178,6 +185,24 @@ public void setNoReconnect(boolean noReconnect) {
         this.noReconnect = noReconnect;
     }
 
+    /**
+     * Returns {@code true} if and only if the socket to the master will have {@link Socket#setKeepAlive(boolean)} set.
+     *
+     * @return {@code true} if and only if the socket to the master will have {@link Socket#setKeepAlive(boolean)} set.
+     */
+    public boolean isKeepAlive() {
+        return keepAlive;
+    }
+
+    /**
+     * Sets the {@link Socket#setKeepAlive(boolean)} to use for the connection to the master.
+     *
+     * @param keepAlive the {@link Socket#setKeepAlive(boolean)} to use for the connection to the master.
+     */
+    public void setKeepAlive(boolean keepAlive) {
+        this.keepAlive = keepAlive;
+    }
+
     public void setCandidateCertificates(List<X509Certificate> candidateCertificates) {
         this.candidateCertificates = candidateCertificates == null
                 ? null
@@ -409,6 +434,7 @@ private Socket connect(@CheckForNull String host, @CheckForNull String port) thr
                 s.connect(targetAddress);
 
                 s.setTcpNoDelay(true); // we'll do buffering by ourselves
+                s.setKeepAlive(keepAlive);
 
                 // set read time out to avoid infinite hang. the time out should be long enough so as not
                 // to interfere with normal operation. the main purpose of this is that when the other peer dies

src/main/java/hudson/remoting/Launcher.java

@@ -189,6 +189,10 @@ public boolean verify(String s, SSLSession sslSession) {
     @Option(name="-noReconnect",usage="Doesn't try to reconnect when a communication fail, and exit instead")
     public boolean noReconnect = false;
 
+    @Option(name = "-noKeepAlive",
+            usage = "Do not open the socket to the master with SO_KEEPALIVE enabled")
+    public boolean noKeepAlive = false;
+
     public static void main(String... args) throws Exception {
         Launcher launcher = new Launcher();
         CmdLineParser parser = new CmdLineParser(launcher);
@@ -229,6 +233,9 @@ public void run() throws Exception {
             if (this.noReconnect) {
                 jnlpArgs.add("-noreconnect");
             }
+            if (this.noKeepAlive) {
+                jnlpArgs.add("-noKeepAlive");
+            }
             try {
                 hudson.remoting.jnlp.Main._main(jnlpArgs.toArray(new String[jnlpArgs.size()]));
             } catch (CmdLineException e) {

src/main/java/hudson/remoting/jnlp/Main.java

@@ -85,6 +85,10 @@ public class Main {
             usage="If the connection ends, don't retry and just exit.")
     public boolean noReconnect = false;
 
+    @Option(name="-noKeepAlive",
+            usage="Do not open the socket to the master with SO_KEEPALIVE enabled")
+    public boolean noKeepAlive = false;
+
     @Option(name = "-cert",
             usage = "Specify additional X.509 encoded PEM certificates to trust when connecting to Jenkins " +
                     "root URLs. If starting with @ then the remainder is assumed to be the name of the " +
@@ -173,6 +177,7 @@ public Engine createEngine() {
         if(jarCache!=null)
             engine.setJarCache(new FileSystemJarCache(jarCache,true));
         engine.setNoReconnect(noReconnect);
+        engine.setKeepAlive(!noKeepAlive);
         if (candidateCertificates != null && !candidateCertificates.isEmpty()) {
             CertificateFactory factory;
             try {