Creating services

One of the main features of the Octue SDK is to allow you to easily create services that can accept questions and return answers. They can run locally on any machine or be deployed to the cloud. Currently:

  • The backend communication between twins uses Google Pub/Sub whether they’re local or deployed

  • The deployment options are Google Cloud Run or Google Dataflow

  • The language of the entrypoint must by python3 (you can call processes using other languages within this though)

Anatomy of an Octue service

An Octue service is defined by the following files (located in the repository root by default).

app.py

This is the entrypoint into your code - read more...

This is where you write your app. The app.py file can contain any valid python, including import and use of any number of external packages or your own subpackages. It has only two requirements:

  1. It must contain exactly one of the octue python app interfaces that serve as an entrypoint to your code. These take a single Analysis instance:

    • Option 1: A function named run with the following signature:

      def run(analysis):
          """A function that uses input and configuration from an ``Analysis`` instance and stores any
          output values and output manifests on it.
      
          :param octue.resources.Analysis analysis:
          :return None:
          """
          ...
      
    • Option 2: A class named App with the following signature:

      class App:
          """A class that takes an ``Analysis`` instance and anything else you like. It can contain any
          methods you like but it must also have a ``run`` method.
      
          :param octue.resources.Analysis analysis:
          :return None:
          """
      
          def __init__(self, analysis, *args, **kwargs):
              self.analysis = analysis
              ...
      
          def run(self):
              """A method that that uses input and configuration from an ``Analysis`` instance and stores
              any output values and output manifests on it.
      
              :return None:
              """
              ...
      
          ...
      
  2. It must access configuration/input data from and store output data on the analysis parameter/attribute:

  • Configuration values: analysis.configuration_values

  • Configuration manifest: analysis.configuration_manifest

  • Input values: analysis.input_values

  • Input manifest: analysis.input_manifest

  • Output values: analysis.output_values

  • Output manifest: analysis.output_manifest

This allows standardised configuration/input/output for services while allowing you to do anything you like with the input data to produce the output data.

twine.json

This is your schema for configuration, input, and output values and manifests - read more...

This file defines your schema for configuration, input, and output values and manifests. Read more here and see an example here.

Dependencies file

A file specifying your app's dependencies - read more...

This is a setup.py file (read more here) or requirements.txt file (read more here) listing all the python packages your app depends on and the version ranges that will work with your app.

octue.yaml

This describes the service configuration - read more...

This file defines the basic structure of your service. It must contain at least:

services:
  - name: my-app

It may also need the following key-value pairs:

  • app_source_path: <path> - if your app.py file is not in the repository root

  • app_configuration_path: <path> - if your app needs an app configuration file that isn’t in the repository root

  • dockerfile_path: <path> - if your app needs a Dockerfile that isn’t in the repository root

All paths should be relative to the repository root. Other valid entries can be found in the ServiceConfiguration constructor.

Warning

Currently, only one service can be defined per repository, but it must still appear as a list item of the “services” key. At some point, it will be possible to define multiple services in one repository.

App configuration file (optional)

An optional app configuration JSON file specifying, for example, any children your app depends on - read more...

If your app needs any configuration, asks questions to any other Octue services, or produces output datafiles/datasets, you will need to provide an app configuration. Currently, this must take the form of a JSON file. It can contain the following keys:

  • configuration_values

  • configuration_manifest

  • children

  • output_location

If an app configuration file is provided, its path must be specified in octue.yaml under the “app_configuration_path” key.

See the AppConfiguration constructor for more information.

Dockerfile (optional)

Provide this if your needs exceed the default Octue Dockerfile - read more...

Octue services run in a Docker container if they are deployed. They can also run this way locally. The SDK provides a default Dockerfile for these purposes that will work for most cases:

However, you may need to write and provide your own Dockerfile if your app requires:

  • Non-python or system dependencies (e.g. openfast, wget)

  • Python dependencies that aren’t installable via pip

  • Private python packages

Here are two examples of a custom Dockerfile that use different base images:

If you do provide one, you must specify its path in octue.yaml under the dockerfile_path key.

As always, if you need help with this, feel free to drop us a message or raise an issue!

Template apps

We’ve created some template apps for you to look at and play around with. We recommend going through them in this order:

  1. The fractal app template - introduces a basic Octue service that returns output values to its parent.

  2. The using-manifests app template - introduces using a manifest of output datasets to return output files to its parent.

  3. The child-services app template - introduces asking questions to child services and using their answers to form an output to return to its parent.

Deploying services automatically

Automated deployment with Octue means:

  • Your service runs in Google Cloud, ready to accept questions from and return answers to other services.

  • You don’t need to do anything to update your deployed service with new code changes - the service simply gets rebuilt and re-deployed each time you push a commit to your main branch, or merge a pull request into it (other branches and deployment strategies are available, but this is the default).

  • Serverless is the default - your service only runs when questions from other services are sent to it, meaning there is no cost to having it deployed but not in use.

To enable automated deployments, contact us so we can create a Google Cloud Build trigger linked to your git repository. This requires no work from you apart from authorising the connection to GitHub (or another git provider).

If you want to deploy services yourself, see here.