For those that get hit with this and the nodes aren't expiring, here is a quick groovy script that will delete all the offline ECS nodes. It should not delete any other slaves under any circumstance.

To run it, just go to the /script directory of your jenkins install.

for (aSlave in hudson.model.Hudson.instance.slaves) {
  println('====================');
  println('Name: ' + aSlave.name);
  if (aSlave.getComputer().toString()  =~ /amazonecs.ECSComputer@/ &&
      aSlave.getComputer().isOffline().toString() == 'true') {
    println('Deleting node!!!!');
    aSlave.getComputer().setTemporarilyOffline(true,null);
    aSlave.getComputer().doDoDelete();
  } else {
    println ('Not an offline ECS node')
  }
}

Hello georgemb, thank you for this fix, it worked.
I noticed also that while i had 200 docker slaves cluttering my jenkins, after running this only once (it deleted all) they started "expiring" now after each build! this is SO weird. It didn't affect the plugin in any way, right? i don't get it, how slave nodes started to behave properly now =)

but what's more interesting, those docker slave nodes were always stuck both in jeknins AND on my ECS instances (stuck as 'stopped' containers) like on this screenshot i just took from my CLI history (before this groovy run):

and now those containers also disappear after build finishes o_O , some kind of voodoo magic it is... i was sure this plugin has a bug (it cluttered instances with containers that not "--rm" themselves). what it might be? i did no changes in my config anywhere .. just tried this script

dima_shmakov The script only affects Jenkins and makes no calls to ECS. ECS will clear out the stopped docker containers it starts after a period. I don't remember what the default is, but it is configurable. The dead containers sticking around shouldn't be a problem, but you can remove them manually if you want.