Closed danielhollas closed 2 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 68.28%. Comparing base (
e2d3d93
) to head (11c4c26
).
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
I think if we go with the zip and decompress solution, the time won't influenced by where it installed.
This may have influenced your testing of your tared-home image, as it would blow up the home size.
But this is true! @superstar54 can you add the change to your PR and check the image size change?
As an alternative to this PR you can merge https://github.com/aiidalab/aiidalab-qe/pull/761 and use uv instead of pip which speeds up the docker build by 1 minute.
If we go with prepare a ready to use home and decompress it for user, we don't need pre-install the dependencies, true? Then I think it is better to install things as what will exactly happened as aiidalab-launch install
.
Hi @danielhollas , thanks for the PR.
it seems that the aiidalab_qe app dependencies were being installed in ~/.local instead of /opt/conda.
Could you give more details on why it is?
@superstar54 please have a look at this PR https://github.com/aiidalab/aiidalab-docker-stack/pull/437
It essentially changed the pip default -- when you run pip install
in the container, it behaves as if you run pip install --user
. So if you don't want that behaviour, you need to specify pip install --no-user
. LMK if you need more info.
If we go with prepare a ready to use home and decompress it for user, we don't need pre-install the dependencies, true? Then I think it is better to install things as what will exactly happened as aiidalab-launch install.
Yeah, you don't need to, though I guess it still helps a bit since it will make home much smaller, and hence untar/decompress should be faster (and the overall image size smaller as well).
Closed in favour of #761
This is an important fix for the Docker build since version v2024.1017 of the aiidalab full-stack image, which added the
--user
argument topip install
by default.@superstar54 it seems that the aiidalab_qe app dependencies were being installed in
~/.local
instead of/opt/conda
. This may have influenced your testing of your tared-home image, as it would blow up the home size.CC @unkcpz
NOTE: As an alternative to this PR you can merge #761 and use uv instead of pip which speeds up the docker build by 1 minute.