Bare Sheep

shepherd.sheep.BareSheep uses a sub-process to run its runner. In particular, it runs shepherd-runner command which is linked to shepherd.runner.runner_entry_point.run().

As mentioned earlier, a sheep communicates with its runner via a socket. The messages are minimal though, in principle, only job_id s are passed through the socket while the data are prepared in a folder.

Configuration

Bare sheep’s config is fairly simple:

bare_sheep:
  port: 9001
  type: bare
  working_directory: examples/docker/emloop_example
  stdout_file: /tmp/bare-shepherd-runner-stdout.txt
  stderr_file: /tmp/bare-shepherd-runner-stderr.txt

Aside from configuration common for all sheep (socket port and sheep type), bare sheep allows to configure:

• working_directory directory from which shepherd-runner command is called

• stdout_file and stderr_file to store the runner outputs

Model Name and Version

Bare sheep allows to run jobs with any model name and version as long as a emloop configuration file can be found in working_directory/model_name/model_version/config.yaml.

For example with working_directory="/var", model_name="my_project/models" and finally model_version empty, the config file is expected to be located in /var/my_project/models/config.yaml.

Usage

Use bare sheep whenever a docker is unavailable or your use case is simple. In other cases, docker sheep is usually recommended.