Automate Sandbox-specific steps for setting up local Dask cluster

[robbibt]

A common science-team workflow is to run several set-up steps to start a local Dask cluster on the Sandbox which do not have to be run on the NCI (see GeoscienceAustralia/dea-notebooks#528).

Some of these steps including dask.config.set, configure_s3_access and perhaps closing previous clients seem like they could be run behind-the-scenes in the Sandbox, removing the need for users to run them every time a notebook is launched on the Sandbox.

This could greatly improve performance for a lot of the code on the Sandbox (e.g. speed improvements of 200% for several of our notebooks), which in turn would improve user experience during demonstrations.

if 'AWS_ACCESS_KEY_ID' in os.environ:
    # configure dashboard link to go over proxy
    dask.config.set({"distributed.dashboard.link":
                 os.environ.get('JUPYTERHUB_SERVICE_PREFIX', '/')+"proxy/{port}/status"})

    # close previous client if any
    client = locals().get('client', None)
    if client is not None:
        client.close()
        del client

    # start up a local cluster  
    client = start_local_dask(mem_safety_margin = '3Gb')

    ## Configure GDAL for s3 access
    configure_s3_access(aws_unsigned=True,  
                        client=client);
else:
    # close previous client if any
    client = locals().get('client', None)
    if client is not None:
        client.close()
        del client

    # start up a local cluster
    client = start_local_dask(mem_safety_margin = '3Gb')

# show the dask cluster settings
display(client)

@Kirill888 @tom-butler @alexgleith

[robbibt]

At this stage I think we're going to wrap the above code in a dea-notebooks function so at least we can more easily use Dask in notebooks without scaring users off with the code block above. @Kirill888 @tom-butler @alexgleith, can you let @cbur24 or myself know if there's any progress on getting some of those Sandbox-specific steps automated in the Sandbox?

[tom-butler]

Most of this should probably be done in the dask config file instead, then a notebook can just include a simple client connection line (similar to datacube)

[robbibt]

I think that sounds great 👍