-
-
Notifications
You must be signed in to change notification settings - Fork 151
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
fix resolve pack method #1533
base: master
Are you sure you want to change the base?
fix resolve pack method #1533
Conversation
mmm maybe we need more thinking: Model eos9f6t to my understanding is not packed with FastAPI (in principle?) but the current fix determines it is. Am I understanding that the "nginx" bit will appear in the history somewhere but not necessarily in the actual "CreatedBy" tagged as "latest"? How these histories are created is a bit confusing to me. This is what I see tagged as latest for the following models:
I do not see a consistency that allows us to determine the packaging manner from this. But I am sure the current "nginx" method is giving problems. Are we sure the eos3b5e is actually well packaged with FastAPI? For the type of information I am seeing it looks more like it is a bentoML model, but it fails. Moreover, this error should be reproducible for everyone but it does not seem the case? I am maybe missing the point on how FastAPI packaged vs BentoML packaged models work. What is the difference on the CREATED BY: ENTRYPOINT and CREATED BY: RUN ? |
Okay definitely looking into this history is no good way of picking the packaging method I tried in another computer and I got an absolute different output for the history of eos3b5e, which does not contain the "nginx" string anywhere not even in the not "latest" images. I am guessing cache plays a role here. This is the history of eos3b5e:
So the history that appears when I print the hist in the function resolve_pack_method_docker is not the same, as elements like this one appear:
|
okay definitely according to GPT the history of an image might differ between the python API and the CLI. Why the behavior is different between computers is a mistery and probably is due to a cached file that does not get removed @Abellegese don't merge the PR as this will not fix the issue system wide probably |
Thank you for taking your time to contribute to Ersilia, just a few checks before we proceed
Description
The resolve pack method for Docker images was not working as the history had several versions some of which contained nginx, which is what was defined to decide between bentoml or fastapi
Changes to be made
I have added an if statement to check only in the latest version
Status
Completed. I have tried to use it with FastAPI packed models that were originally packed with BentoML (eos3b5e) and
models that have been packed with BentoML (eos9f6t) for example. Works in both situations.
To do
None
Is this pull request related to any open issue? If yes, replace issueID below with the issue ID
Related to #1509