Prototype implementation of durable functions backend.

[Bubblyworld]

See #123. Implements an azure durable functions backend for SAME, which can be targeted using same run -t functions [...]. Requires credentials to be provided either using environment variables (see azure documentation on service principals), or using cached credentials from running az login (more useful if you have a personal azure account).

The current set of features are:

• Automatically provisions the necessary resources on azure when deploying, and does this idempotently.
• Deploys notebooks as JSON blobs to a configured HTTP endpoint that orchestrates execution.
• Returns a link to a HTTP endpoint that can be queried for execution status, results etc.

What is not currently supported:

• Multistep notebooks.
• Notebooks with requirements.
• Notebooks with custom docker images.

I've left TODO markers around the codebase to indicate where the missing features can be implemented. Should mostly be quite straightforward, with the exception of custom docker images, as a durable function runs in a single pre-configured container.

What a journey! Took much longer than expected, documentation for the python azure SDK is lacking, to say the least.

[Bubblyworld]

The test suite can be run against durable functions with the --functions flag:

FUNCTIONS_SUBSCRIPTION_ID="<azure subscription ID>" pytest --functions

If the functionapp has already been provisioned, the FUNCTIONS_SKIP_PROVISIONING="true" environment variable can also be set to speed the tests up significantly. An example of running a SAME notebook against the functions backend:

The resulting URL can be cURLed to obtain execution status and results:

[github-actions[bot]]

Unit Test Results

  1 files    1 suites   13m 43s :stopwatch: 95 tests 66 :heavy_check_mark: 29 :zzz: 0 :x:

Results for commit 90a08f84.