Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Using unified conda-lock file in postBuild for reproducible conda environment #14

Draft
wants to merge 31 commits into
base: main
Choose a base branch
from

Conversation

weiji14
Copy link
Member

@weiji14 weiji14 commented Nov 29, 2022

Towards better reproducibility of the conda environment. Done by having a custom conda-lock command in postBuild, rather than letting repo2docker build directly from an enviroment.yml file. Suggestion from #11 (comment)

TODO:

  • Rename environment.yml file to environment-cryocloud.yml (so that Binder doesn't use it. Install from lockfile happens in postBuild)
  • Change conda-lock command from generating explicit lockfiles to unified lockfiles
  • Implement postBuild bash script
  • Test to ensure things work (Test locking unified condalock attempt 5 weiji14/hub-image#8)

Addresses #11. Closes #15.

So that repo2docker doesn't use it to build the docker image.
@weiji14 weiji14 self-assigned this Nov 29, 2022
@github-actions
Copy link

Binder 👈 Test this PR on Binder

Minor version updates to fix `mamba` solver.
@weiji14
Copy link
Member Author

weiji14 commented Nov 29, 2022

/condalock

Handle both conda and pip dependencies by changing from the old explicit conda-lock (<1.0) format to the newer unified format. Deleting conda-linux-64.lock and replacing it with a placeholder conda-lock.yml for now.
@weiji14
Copy link
Member Author

weiji14 commented Nov 29, 2022

/condalock

weiji14 and others added 10 commits November 29, 2022 15:50
Need a proper placeholder conda-lock.yml to prevent `AttributeError: 'NoneType' object has no attribute 'pop'` when running `conda-lock lock`.
The `--file` option doesn't actually work, even though it's documented like so.
Update instructions on how to add or update dependencies in the cryocloud docker image. Specifically, the part where `conda-lock` is used for a fully reproducible conda environment.
Needed to avoid `Warning: found pip deps, but conda-lock was installed without pypi support. pip dependencies will not be included in the lock file`. See https://conda-incubator.github.io/conda-lock/getting_started.
No more explicit conda-linux-64.lock, just the unified conda-lock.yml.
@weiji14 weiji14 marked this pull request as ready for review November 29, 2022 22:09
Follow the default environment name on the JupyterHub rather than the conda environment's name.
@@ -42,12 +42,12 @@ jobs:

# Install conda-lock library
- name: Install conda-lock
run: pip install conda-lock
run: pip install conda-lock[pip_support]==1.0.5
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Using conda-lock 1.0.5 instead of the latest 1.2.1 to keep the relative file paths intact. See #15.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, might need to upgrade to conda-lock 1.2.1 actually, because it contains important bugfixes related to pip dependency hashes. E.g. see https://github.com/CryoInTheCloud/hub-image/actions/runs/3578432352/jobs/6018581680#step:5:4747

  ERROR:root:ERROR: THESE PACKAGES DO NOT MATCH THE HASHES FROM THE REQUIREMENTS FILE. If you have updated the package versions, please update the hashes. Otherwise, examine the package contents carefully; someone may have tampered with them.
  ERROR:root:    fastjsonschema@ https://files.pythonhosted.org/packages/e4/be/cf1b876348070a23cb0c3ebfee7a452ad3a91b07b456dade3bd514656009/fastjsonschema-2.16.2-py3-none-any.whl#md5=None from https://files.pythonhosted.org/packages/e4/be/cf1b876348070a23cb0c3ebfee7a452ad3a91b07b456dade3bd514656009/fastjsonschema-2.16.2-py3-none-any.whl#md5=None (from -r /tmp/tmpn_q_dp50 (line 1)):
  ERROR:root:        Expected md5 None
  ERROR:root:             Got        fff77553c8612b49c8782a5233541199
  ERROR:root:
  ERROR:root:
  ERROR:root:ERROR conda.cli.main_run:execute(49): `conda run pip install --no-deps -r /tmp/tmpn_q_dp50` failed. (See above for error)
  Traceback (most recent call last):
    File "/srv/conda/envs/notebook/bin/conda-lock", line 8, in <module>
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1130, in __call__
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1055, in main
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1657, in invoke
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1404, in invoke
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 760, in invoke
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/decorators.py", line 26, in new_func
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/conda_lock/conda_lock.py", line 1268, in install
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/conda_lock/conda_lock.py", line 219, in do_conda_install
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/conda_lock/invoke_conda.py", line 131, in _invoke_conda
  poetry.utils._compat.CalledProcessError: Command '['/srv/conda/bin/mamba', 'run', '--name', 'notebook', 'pip', 'install', '--no-deps', '-r', '/tmp/tmpn_q_dp50']' returned non-zero exit status 1.
  Removing intermediate container 4d439bce016f
  The command '/bin/sh -c ./postBuild' returned a non-zero code: 1

#!/bin/bash
set -eux
pip install conda-lock[pip_support]==1.0.5
conda-lock install --mamba --name notebook conda-lock.yml
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure if it matters, but I've set the conda environment name as notebook here instead of cryocloud, because that's the default conda environment name on the CryoCloud JupyterHub.

weiji14 and others added 3 commits November 29, 2022 17:37
Also drop the pip_support extras since poetry has been vendored in conda-lock 1.2.0.
Using GitHub Actions hosted runner filepath.
postBuild Outdated Show resolved Hide resolved
postBuild Outdated Show resolved Hide resolved
Copy link
Contributor

@yuvipanda yuvipanda left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

couple comments, otherwise this looks great!

Co-Authored-By: Yuvi Panda <yuvipanda@gmail.com>
@weiji14
Copy link
Member Author

weiji14 commented Nov 29, 2022

Hitting into this bug with pywinpty, see conda/conda-lock#236. Error from https://github.com/CryoInTheCloud/hub-image/actions/runs/3578701513/jobs/6019136464#step:5:4686

  INFO:root:Collecting pywinpty@ https://files.pythonhosted.org/packages/22/77/59645ee028f410ef1d582fc2923d2cb61016ed38e6ae5022f405227ca2ad/pywinpty-2.0.9.tar.gz#sha256=01b6400dd79212f50a2f01af1c65b781290ff39610853db99bf03962eb9a615f
  INFO:root:  Downloading pywinpty-2.0.9.tar.gz (26 kB)
  INFO:root:  Installing build dependencies: started
  INFO:root:  Installing build dependencies: finished with status 'done'
  INFO:root:  Getting requirements to build wheel: started
  INFO:root:  Getting requirements to build wheel: finished with status 'done'
  INFO:root:  Preparing metadata (pyproject.toml): started
  INFO:root:  Preparing metadata (pyproject.toml): finished with status 'error'
  INFO:root:
  ERROR:root:  error: subprocess-exited-with-error
  ERROR:root:
  ERROR:root:  × Preparing metadata (pyproject.toml) did not run successfully.
  ERROR:root:  │ exit code: 1
  ERROR:root:  ╰─> [6 lines of output]
  ERROR:root:
  ERROR:root:      Cargo, the Rust package manager, is not installed or is not on PATH.
  ERROR:root:      This package requires Rust and Cargo to compile extensions. Install it through
  ERROR:root:      the system's package manager or via https://rustup.rs/
  ERROR:root:
  ERROR:root:      Checking for Rust toolchain....
  ERROR:root:      [end of output]
  ERROR:root:
  ERROR:root:  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR:root:error: metadata-generation-failed
  ERROR:root:
  ERROR:root:× Encountered error while generating package metadata.
  ERROR:root:╰─> See above for output.
  ERROR:root:
  ERROR:root:note: This is an issue with the package mentioned above, not pip.
  ERROR:root:hint: See above for details.
  ERROR:root:
  ERROR:root:ERROR conda.cli.main_run:execute(49): `conda run pip install --no-deps -r /tmp/tmpp9s90hbc` failed. (See above for error)
  Traceback (most recent call last):
    File "/srv/conda/envs/notebook/bin/conda-lock", line 8, in <module>
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1130, in __call__
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1055, in main
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1657, in invoke
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 1404, in invoke
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/core.py", line 760, in invoke
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/click/decorators.py", line 26, in new_func
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/conda_lock/conda_lock.py", line 1288, in install
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/conda_lock/conda_lock.py", line 220, in do_conda_install
    File "/srv/conda/envs/notebook/lib/python3.7/site-packages/conda_lock/invoke_conda.py", line 131, in _invoke_conda
  conda_lock._vendor.poetry.utils._compat.CalledProcessError: Command '['/srv/conda/bin/mamba', 'run', '--name', 'notebook', 'pip', 'install', '--no-deps', '-r', '/tmp/tmpp9s90hbc']' returned non-zero exit status 1.
  Removing intermediate container f13e567c749f
  The command '/bin/sh -c ./postBuild' returned a non-zero code: 1

This might take a while to resolve on conda-lock, and I'm not sure if there's a good workaround. I think it'll be best if I cherry-pick 90fe9f6 into a separate PR so that #13 can proceed.

@tsnow03 tsnow03 marked this pull request as ready for review November 30, 2022 22:16
Add back in the 'start'` file to `.github/workflows/build.yaml`.
Replaced by conda-linux.yml.
@weiji14 weiji14 marked this pull request as draft December 4, 2022 04:01
@weiji14
Copy link
Member Author

weiji14 commented Dec 4, 2022

Marking as draft again as the bug at #14 (comment) isn't fixed yet, waiting on conda-forge/staged-recipes#21369 🤞 (which is itself waiting on a few other things).

@yuvipanda
Copy link
Contributor

conda-forge/staged-recipes#21369 might take a while I think :D Does this mean @weiji14 this repo can only install from conda and can't install from pip?

@weiji14
Copy link
Member Author

weiji14 commented Dec 5, 2022

conda-forge/staged-recipes#21369 might take a while I think :D Does this mean @weiji14 this repo can only install from conda and can't install from pip?

Yes, only via mamba/conda. Using conda-lock.yml won't work unless we get jupyter-desktop-server on conda-forge, or the issue with winpty gets resolved in conda-lock.

@yuvipanda
Copy link
Contributor

Hmm, that seems a major limitation going forward honestly :| PyPI will always have things that won't be on conda-forge I think. Do you know if conda-lock might resolve this, or is it a non-goal?

@weiji14
Copy link
Member Author

weiji14 commented Dec 5, 2022

Hmm, that seems a major limitation going forward honestly :| PyPI will always have things that won't be on conda-forge I think. Do you know if conda-lock might resolve this, or is it a non-goal?

The issue is tracked in conda/conda-lock#236, and actually, jupyter-desktop-server is a special case because it indirectly depends on winpty which is Windows only. Most other PyPI packages won't have this issue, so I still think it's ok.

@yuvipanda
Copy link
Contributor

AAAAAAH, apologies @weiji14 for not following links - and thank you so much for clarifying this is desktop-server specific!

@weiji14 weiji14 marked this pull request as ready for review May 28, 2023 22:32
@weiji14
Copy link
Member Author

weiji14 commented May 28, 2023

Ready for review @yuvipanda! I've updated to conda-lock=2.0.0 which fixes the Windows-only pywinpty issue (thanks to @scottyhq's patch in conda/conda-lock#295), so we don't need to wait for a jupyter-desktop-server conda-forge package anymore 🎉 See weiji14@ec088de for an example of /condalock working successfully with the new update.

Once this is merged, and tested a few times, we should be able to move ahead with the /condalock GitHub Action work as mentioned in #18.

weiji14 and others added 2 commits June 10, 2023 14:01
Co-authored-by: Matt Fisher <mfisher87@gmail.com>
Conflicts:
	conda-linux-64.lock
@weiji14 weiji14 marked this pull request as draft August 29, 2023 01:45
@weiji14
Copy link
Member Author

weiji14 commented Aug 15, 2024

Note to self. A 'conda-lock.yml' file was added in #122 already, though the builds which rely on repo2docker are technically still using 'environment.yml' so not 100% reproducible yet

- name: Build and push the image to quay.io
uses: jupyterhub/repo2docker-action@master

Maybe wait for repo2docker to support conda-lock natively (jupyterhub/repo2docker#1312), and then the 'postBuild' hack here can go away.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

update conda-lock version after updates
4 participants