We run build nodes via Docker and the Swarm Plugin. After a while defunct processes start to pile up on the nodes:

1000 9386 9371 1 Dec18 ? 00:10:49 java -jar /usr/share/jenkins/swarm-client-3.15-jar-with-dependencies.jar -fsroot /var/jenkins-node_home -master https://jenkins.cosmos.local -username XXX -password XXX -executors 10 -mode exclusive -labels linux basic build -name basic-node-dc -disableSslVerification -description Basic node

[root@XXX:/root]# ps -ef | grep defu 
1000 2489 9386 0 Dec18 ? 00:00:00 [ssh-agent] <defunct> 
1000 2514 9386 0 Dec18 ? 00:00:00 [ssh-agent] <defunct> 
1000 2544 9386 0 Dec18 ? 00:00:00 [ssh-agent] <defunct> 
1000 2618 9386 0 Dec18 ? 00:00:00 [ssh-agent] <defunct> 

...

We run ssh-agent often through many scripted pipelines so it is hard to trace it down to a specific Pipeline, but this behavior shouldn't occur to begin with.