Connect to dask-scheduler from local workstation (maybe update documentation?)

[manuel-rhdt]

I connect to dask-scheduler from my local workstation to submit my computations (using a ssh tunnel).

I have had other people tell me a similar thing (third bullet point of #186 (comment)). If you manage to make that work, please let us know (ideally in a separate issue).

Originally posted by @lesteve in #186 (comment)

This helps circumvent problems where the login node of a cluster has very limited computing power and where you are not allowed to submit jobs from interactive nodes.

The idea is that the login node runs dask-scheduler and we submit the computations to it from a local workstation. The steps to do this:

1. start ipython on the login node and start the cluster, e.g.

from dask_jobqueue import PBSCluster 
cluster = PBSCluster(…)
cluster.scale(100)  

2. check the address of the cluster

In [2]: cluster                                                 
Out[2]: PBSCluster('tcp://192.168.57.5:43704', workers=100, threads=100, memory=250GB)

3. Taking note of the port of the scheduler (43704), now on the local machine setup an ssh tunnel to the cluster (and the dashboard)

ssh -N -L 8786:localhost:43704 -L 8787:localhost:8787 login.cluster

4. You should be able to access the dashboard on the local machine (localhost:8787)

5. Setup the Client on your local machine as follows:

from dask.distributed import Client
client = Client("tcp://localhost:8786")

Now you should be able to submit computations to the cluster from your local machine.

[guillaumeeb]

Thanks @manuel-rhdt, would you mind proposing a PR to add this in a page from the detailed use section of the docs?