ssh cluster does not close

Hi,

when starting a sshcluster via:

from distributed.deploy.ssh import SSHCluster

def start_cluster():
    hosts = ['localhost']
    nthreads = 1
    nprocs = 2
    c = SSHCluster(hosts * (nprocs+1),
                   connect_options={"known_hosts": None},
                   worker_options={"nthreads": nthreads, },
                   scheduler_options={"port": 0, "dashboard_address": ":8787"},
                   )
    c.scale(nprocs)
    c.close()

if __name__ == '__main__':
    start_cluster()

the scheduler and the worker are still alive even if the calling python process is finished. The dashboard is reachable and the python processes are shown in “top”.

Is there another way to close the ssh cluster?

Using Linux: distributed 2.9.1 dask 2.9.1 python 3.6.9

[condoratberlin]

It seems to fixed in OpenSSH version 7.9 as described in the release nodes.

[Keou0007]

No immediate happiness with the patch, I’ll post my findings on the PR.

I did figure out a workaround for this issue that works for me for now. I followed partially the instructions here to run a second instance of ssh-server on a different port and configured the init script to launch the conda-forge version of sshd. I did need to modify the settings in anaconda/etc/sshd_config a bit to make it work without complaining.

I then provide that port number in the connect_options dict when I launch the SSHCluster and it connects to the newer version of the ssh server and works as expected.