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
[MRG] install notebook in its own env #651
Conversation
leave the conda root env alone adds conda activation to profile.d and ensures that we start with a login shell via the ENTRYPOINT
I think this is going to result in a performance improvement, since it skips a whole install step (the Python switch), and creates a new env instead of updating the root env. The result is an ENTRYPOINT that ensures we launch with a fully configured login shell (profile.d, bashrc are sourced, unlike before), including the shell integration in #567. This also means that our notebook env is fully activated, in case there are any conda activation hooks in that env. This will also facilitate #637 because we don't care what version Python is in the conda root env |
rather than replacing our entrypoint
in tests/conda/simple use pytest for better error reporting
- conda is not importable in the frontmost python - sys.executable is in $NB_PYTHON_PREFIX (this was true before, but it's value has changed)
Additionally, this moves the handling of start_script down a level, so that our ENTRYPOINT is always called. The behavior should be the same. |
like other things, this was always the right thing to do, but more important now that notebook prefix is not the same as CONDA_DIR
I think some of the failures are because we are still using conda 4.5 and not 4.6. For example the At least for https://travis-ci.org/jupyter/repo2docker/jobs/524519120#L931-L940 |
@betatim good call. I merged in #637 and bumped conda up to 4.6.14 (there is now a miniconda 4.6.14) with a refreeze. This should mean significant performance improvements because we skip two |
I re-added the step to make the start script executable as I think that was the cause for some of the tests to fail. Maybe you can squash the two commits into one (I used the webinterface :-/) |
I think https://github.com/jupyter/repo2docker/blob/bc9554e18cead7336cb66daee0435a3d63d701b4/tests/julia/julia_version-default/verify#L20 is the line that fails in the Julia tests. |
I think this is basically ready to be merged (I will fix the broken test now) and then we can start a new PR/rebase #657. What do you think @yuvipanda? |
I propose we merge this now. All tests pass, it gets us unblocked on conda versions, and one step closer to always splitting notebook server and user's kernel env. |
I’m happy to merge this now. Sorry I didn’t finish it earlier. |
No worries. We are IMHO doing a good job of not having to be in a hurry. Thanks for sorting us out with a nice new codna env setup! |
[MRG] install notebook in its own env
leave the conda root env alone
adds conda activation to profile.d and ensures that we start with a login shell via the ENTRYPOINT
closes #567 by adding an ENTRYPOINT that enables conda integration
cc @betatim