Can't access dashboard

Apologies for bringing this up again as it has been discussed in https://github.com/dask/dask-jobqueue/issues/71 and https://github.com/dask/dask-jobqueue/issues/80. The discussions in those haven’t helped. Hopefully this is a quick fix and I can show my colleagues how cool the LFSCluster is 😃

I tried the example from pangeo (http://pangeo-data.org/setup_guides/hpc.html#deploy-option-1-jupyter-dask-jobqueue).

Start a jupyter notebook on the HPC as: jupyter notebook --no-browser --port 8888. Then on my laptop I try: ssh -N -L 8888:localhost:8888 -L 8787:localhost:8787 [e-mail] This gives:

channel 4: open failed: connect failed: Connection refused
channel 5: open failed: connect failed: Connection refused

In this case I can’t access http://localhost:8888 let alone the dashboard. I’m not quite sure why this happens. I’ve had success in the past with tunneling notebooks as ssh -N -L 8888:localhost:8888 [e-mail]. When I haven’t even been able to connect to http://localhost:8888 restarting my laptop seems to work (i’m sure there is a less drastic fix).

Next I tried the example in the docs https://dask-jobqueue.readthedocs.io/en/latest/examples.html?highlight=tunnel#viewing-dask-dashboard

Start ipython on the HPC and inspect the client:

Then on my laptop use this address for tunneling: ssh -fN rxb826@10.10.0.14 -L 8787:localhost:8787. Then when I try to connect to http://localhost:8787 I see:

It seems I can’t ping this address

I sent this to my IT dept as well. Will update with any info at that end.

[raybellwaves]

http://pangeo-data.org/setup_guides/hpc.html#deploy-option-1-jupyter-dask-jobqueue did have the solution after all.

I would open ipython and inspect the client

In [1]: from dask_jobqueue import LSFCluster

In [2]: from dask.distributed import Client

In [3]: cluster = LSFCluster(cores=2, memory='4 GB', walltime='00:02')

In [4]: client = Client(cluster)

In [5]: client
Out[5]: <Client: scheduler='tcp://10.10.0.14:42151' processes=0 cores=0>

Next I would use that ip to start my notebook jupyter notebook --no-browser --ip=10.10.0.14 --port=8888

Then on my laptop tunnel as ssh -N -L 8888:10.10.0.14:8888 [e-mail] and open http://localhost:8888. Work though the notebook and inspect the client. The dashboard will most likely by port 8787 but sometimes it’ll be a five digit port number e.g. 54611. Open another terminal and tunnel the dashboard as ssh -N -L 54611:10.10.0.14:54611 [e-mail]. Lastly, open http://localhost:54611 on my laptop.

I can make a change to the docs if agreed https://dask-jobqueue.readthedocs.io/en/latest/examples.html?highlight=dashboard#viewing-dask-dashboard

[guillaumeeb]

It is true that this part of the docs is a little misleading with the localhost everywhere, any improvement welcome!