Skip to main content

3. Starting agents

Repositories are hosted by agents - each agent can have its own package dependencies, and be deployed within your infrastructure as needed - for example one agent could be deployed on an on-premise bare-metal server with a GPU, and another agent could be deployed as a Docker image on an auto-scaling cloud cluster.

Primarily, an agent is a process that's responsible for executing the code required by your workflow. Additionally, it will:

  1. Register a manifest describing the targets (workflows/tasks/etc) that it supports, to the orchestrator.
  2. Listen for commands from the orchestrator.
  3. Invoke and monitor executions of operations (in forked sub-processes).
  4. Report the status (including results, errors, etc.) of executions back to the orchestrator.

Install​

The agent, library and CLI can be installed from PyPI - e.g., with pip:

pip install coflux

Run the command above, or install the package as you prefer.

Initialise​

First, use the configure command to populate a configuration. This isn't necessary, but avoids having to specify configuration manually in the following commands. Run the following command:

coflux configure

You will be prompted to enter the host (localhost:7777), the project ID, and the environment name.

Run​

Now the agent can be started. Run the following command:

coflux agent.run hello.py

In the web UI you will be able to see your workflow.

Next, let's initiate a run...