pip install --upgrade randomly skips upgrading a dependency leading to dependency errors #8218
Labels
C: dependency resolution
About choosing which dependencies to install
Comments
triage-new-issues
bot
added
the
S: needs triage
|
Related: #8115 |
|
@uranusjr Do you think there is a way to stabilize pip install --upgrade behavior? |
|
@Ark-kun The problem should be fixed in the new resolver - can you confirm if |
pradyunsg
added
C: dependency resolution
triage-new-issues
bot
removed
the
S: needs triage
Ark-kun
added a commit
to Ark-kun/pipelines
that referenced
this issue
May 21, 2020
Ark-kun
added a commit
to Ark-kun/pipelines
that referenced
this issue
May 21, 2020
I've checked and the new resolver just fails. This is understandable, since some other dependencies (not the unstable one) are incompatible. https://travis-ci.com/github/kubeflow/pipelines/jobs/338350956#L2500 |
no-response
bot
removed
the
S: awaiting response
|
Is it possible to make pip be deterministic at least? There is a ~66% chance that this would lead to success for us. |
|
@pfmoore I just ran into this same issue, except the upgrade was less likely to happen (most of our dependencies depend on numpy). Locally, |
|
@Ark-kun Could you try it again with the new resolver |
|
This issue has been automatically closed because there has been no response to our request for more information from the original author. With only the information that is currently in the issue, we don't have enough information to take action. Please reach out if you have or find the answers we need so that we can investigate further. |
pradyunsg
removed
the
S: awaiting response
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Environment
Description
We have presubmit tests in Travis that test our python SDK and some other projects that depend on our SDK.
Recently the tests started getting flaky. Tests for changes that do not change the code (e.g. change README) sometimes pass and sometimes fail. I've compared the successful and failing runs and noticed what seems to be a randomly-triggered bug in
pip install --upgradebehavior.When installing a wheel we want to test, some of its dependencies depend on the
numpy. Most depend onnumpy<2,>=1.16, but one dependency also supports older versionsnumpy<2,>=1.14.3. The environment where the installation runs hasnumpy-1.15.4- enough for one dependency, fut too old for others.We observe that
pip install --upgraderandomly skips upgrading numpy which leads to broken dependencies that pip install reports in the end.Expected behavior
I expect
pip install --upgradehave deterministic behavior.I expect that
pip install --upgrade, when installing multiple dependencies requiring a different minimum version of some package, will eventually install a package version which satisfies all min version requirements.I understand that pip does not do full dependency resolution. But given
A requires p>=1; B requires p>=2; C requires p>=3;I expectp>=3to be installed eventually.Actual behavior
We observe what amounts to pip only checking a package requirement once (randomly selected) and if the requirement is satisfied it ignores other requirements.
I.e. sometimes it would work as expected:
and sometimes it would
How to Reproduce
Working on PoC.... It's a bit complicated since it involves
git clone,setup.py bdist_wheeland thenpip install.Output
Works:
Does not work:
The full failed log can be currently found at https://travis-ci.com/github/kubeflow/pipelines/jobs/331132675, but it will go away soon.
The text was updated successfully, but these errors were encountered: