Jenkins installs "ping thread" on every remoting connection, such as master/slave connections, regardless of its transport mechanism (such as SSH, JNLP, etc.) The lower level of the Jenkins remoting protocol is a message oriented protocol, and a ping thread periodically sends a ping message that the receiving end will reply. The ping thread measures the time it takes for the reply to arrive, and if it's taking excessive time (currently 4 minutes and configurable), then it assumes that the connection was lost and initiates the formal close down.
This is to avoid an infinite hang, as some of the failure modes in network cannot be detected otherwise. The timeout is also set to a long enough value so that a temporary surge in the load or a long garbage collection pause will not trip off the close down.
Ping thread is installed on both ends; each side pings the other and tries to detect the problem from their own sides.
The ping thread time out is reported through java.util.logging. In addition, the master will also report this exception in the slave launch log. Note that some slave launchers, most notably SSH slaves, writes all stdout/stderr outputs from the slave JVM into this same log file, so you need to be careful. See JENKINS-25695.
Disabling ping thread
Sometimes, for example to diagnose the slave connection loss problem, you may want to disable the ping thread. This needs to be done in two places.
To disable the master from pinging slaves, the system property hudson.slaves.ChannelPinger.pingInterval should be set on the master JVM to -1. You can also change the value in memory for a running Jenkins, if you don't want to restart Jenkins (Deprecated since 2.37):
Jenkins.instance.injector.getInstance(hudson.slaves.ChannelPinger.class).@pingInterval = -1
Since 2.37, set pingIntervalSeconds and pingTimeoutSeconds on the master JVM to -1:
Jenkins.instance.injector.getInstance(hudson.slaves.ChannelPinger.class).@pingIntervalSeconds = -1 Jenkins.instance.injector.getInstance(hudson.slaves.ChannelPinger.class).@pingTimeoutSeconds = -1
The above will only affect newly connected slaves. Existing connected slaves will continue running pings.
To disable slaves from pinging the master, the system property -Dhudson.remoting.Launcher.pingIntervalSec=-1 needs to be set to slaves. How to do this depends on the launcher.