You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I searched open requests and couldn't find a duplicate
What is the idea?
Incremental Packaging Support in conda-pack
One new command that can merge two conda environments with the same Python version
Why is this needed?
Scenario: In a unified container environment, a user uses a pre-built conda environment for development tasks, and for a variety of reasons, the user may make changes to packages in the current conda environment (add/remove/update...) . We want to deploy the user's code with the modified environment.
Status quo: it is easy to package existing environments using conda-pack, but the problem is that when users are working on scientific computing tasks, they often introduce a large number of bulky libraries whose conda env may be tens of gigabytes in size, and transferring this environment to the deployment environment (which may have multiple nodes), or packaging this environment as a container image significantly raises the storage usage.
Benefit: With incremental packages, we only need to transfer the incremental packages, not the entire conda environment, and we can directly update the conda environment inside the container, which is similar to git merge!
Alternative: Download packages over the network, but internet connections may be regulated in some organizations. Save the user's image (docker commit), but this may contain more useless information.
What should happen?
Input:
List of excluded Conda and pip packages, including their versions. (In the above scenario, it is the package and its version of the container environment that is provided to the user)
Output:
A archive containing packages that are not in the excluede, or are different from the exclude version
Incremental migration: Remove old and install new
Remove duplicate packages from archive in new environment
Extract the packages from archive to the new environment
Exception:
When the two python versions are not the same, raise an error
I understand that this goal may be a bit different from the original goal of the conda pack, and I've found that a lot of the code in the conda pack can be reused.
I'd like to know if conda pack is interested in adding this feature, and if so, I can make a pr for it.
The text was updated successfully, but these errors were encountered:
Checklist
What is the idea?
conda-pack
Why is this needed?
Scenario: In a unified container environment, a user uses a pre-built conda environment for development tasks, and for a variety of reasons, the user may make changes to packages in the current conda environment (add/remove/update...) . We want to deploy the user's code with the modified environment.
Status quo: it is easy to package existing environments using conda-pack, but the problem is that when users are working on scientific computing tasks, they often introduce a large number of bulky libraries whose conda env may be tens of gigabytes in size, and transferring this environment to the deployment environment (which may have multiple nodes), or packaging this environment as a container image significantly raises the storage usage.
Benefit: With incremental packages, we only need to transfer the incremental packages, not the entire conda environment, and we can directly update the conda environment inside the container, which is similar to
git merge
!Alternative: Download packages over the network, but internet connections may be regulated in some organizations. Save the user's image (
docker commit
), but this may contain more useless information.What should happen?
Input:
Output:
Incremental migration: Remove old and install new
Exception:
Example:
Packaging an incremental archive
Incremental installation
Additional Context
I understand that this goal may be a bit different from the original goal of the conda pack, and I've found that a lot of the code in the conda pack can be reused.
I'd like to know if conda pack is interested in adding this feature, and if so, I can make a pr for it.
The text was updated successfully, but these errors were encountered: