Possibilities to create task which won't be killable before a server reboot :

Easy way to reproduce.

1-) Create a task with a long loop
2-) While task is in the loop, blue screen the slave
3-) Go on another computer and try to abort the task while the slave is down, at least 2 times
4-) Reboot slave.

Note that if step 3 is not done, problem seems to not be happening.

Now task won't be able to be killed. May dies after 20 minutes!
If you want a way to have the task to now even die after 20 minutes, crash computer doing something else, like
1-) sending email after a failure
2-) trying to build artifacts
Then after 2 hours it still was not closed.

Joined the result of the task as picture.