Investigate reuse strategies for docker image

[ckunki]

Potential use cases

UC-1

• notebook developer Nadine works on creating a new Jupyter notebook or updating an existing one
• Nadine wants to use new libraries that are not available yet in the latest release on docker-hub
• Nadine therefor wants to build a private Docker image from the branch Nadine is currently working on

UC-2

• As image creation currently (Nov 2023) takes around 7 minutes Nadine wants to reuse the image in follow-up usage

UC-3

• Nadine changed a file or dependency that requires to re-create the Docker image taking the change into account

[ckunki]

Using a stripped playbook I measured a total duration of 35 seconds.
The main runtime (> 30 s) is used for copying ~ 1MB files into the Docker container.

When using the full playbook the

• duration before ansible starts is negligible
  • 13:19:02.553448520 Setup DSS Docker Container
  • 13:20:06.942173827 Install Poetry
  • 13:21:02.421321996 Install Jupyter
  • 13:21:33.782174510 Install JupyterLab and its dependencies
  • 13:22:04.042733300 Install dependencies used in Jupyterlab
  • 13:22:49.754936638 Copy notebook content
  • 13:23:01.747934461 Install Docker
  • 13:23:35.444312396 docker : Add Docker GPG apt Key
  • 13:23:51.388766915 docker : Add Docker Repository
  • 13:24:02.429195331 docker : Update apt and install docker-ce
  • 13:24:52.802106986 Ansible tasks finished
  • 13:25:07.204318968 stopping container
  • 13:25:12.117735411 PASSED
  • 13:25:23.652135227 End of test

[ckunki]

Comments from @tkilias

• Installation of poetry takes quite long and I think we don't need poetry
• Yes, APT and PIP installs take a while but there are only limited things we could do about it
• In a Docker Container we don't need systemd and motd setup as they probably won't work anyway

[ckunki]

See https://stackoverflow.com/questions/27985334/why-is-copying-a-directory-with-ansible-so-slow:
TLDR: use synchronize instead of copy.

[ckunki]

See also https://docs.pytest.org/en/7.4.x/how-to/capture-stdout-stderr.html

def test_disabling_capturing(capsys):
    print("this output is captured")
    with capsys.disabled():
        print("output not captured, going directly to sys.stdout")
    print("this output is also captured")

You tried to access the function scoped fixture capsys with a session scoped request object, involved factories:

[ckunki]

I used

- name: Copy notebook content
  ansible.builtin.synchronize:
    src: "roles/jupyter/files/notebook/"
    dest: /root/notebooks
    rsync_opts:
      - "--chmod=0644"

And got error message

protocol version mismatch -- is your shell clean?
(see the rsync man page for an explanation)
rsyncerror: protocol incompatibility (code 2) at compat.c(178) [sender=3.1.3]

After adding rsync to the Docker Container and replacing ansible.builtin.copy by ansible.builtin.synchronize the duration decreased from ~ 24 seconds to < 1 second!
Great! 🎉

Unfortunately installing rsync itself takes 31 seconds: 🙁

[ckunki]

Capturing stdout in @pytest.fixture(scope="session") only works then pytest is called with -o log_cli=true -o log_cli_level=INFO.

But when these cli options are provided capturing is not required anymore, as pytest will log ansible output anyway.

[ckunki]

For reusing a Docker Image in pytest DSS added support for CLI option --dss-docker-image in #69