Add standard mapping option for environment variables.

In our Pangeo deployments, we have a need for setting an environment variable on the workers based on user input. This seems like the kind of thing that would be useful for all backends so I’m raising this here while we’ll implement it on our own in Pangeo for now.

1. A widget / traitlet for accepting input from the user in the form of a mapping from {name: value}.
2. A standard dask-gateway config for this option for all backends supported by dask-gateway, which is then forwarded to the appropriate location.

The main downside I see with 2 is that this would make two ways to specify environment variables for each backend. The “dask-gateway” way, and the backend-native way. For kubernetes that would be

gateway:
  backend:
    worker:
      extraContainerConfig:
        env:
          - name: name
            value: value

Is that problematic?

cc @rsignell-usgs and @scottyhq.

[jcrist]

I’ve added a mapping option in #290, will be part of the next release.

[jcrist]

Is it the case that dask-gateway necessarily restricts user-defined dask config to whatever subset of the full dask configuration options are exposed from here: https://gateway.dask.org/api-server.html#cluster-backends ?

We don’t do anything with preload scripts currently.

Also, note that configuration set locally (using dask.config.set) isn’t generally picked up by a dask cluster started via any of dask’s clusters tools. If you’re running a cluster that starts the scheduler locally, then the scheduler and client will see your configuration, but the workers will not. Dask-gateway doesn’t run the scheduler locally, so only the Client will see your local dask configuration. All options on dask-gateway clusters must be either set by administrators, or exposed via cluster options.