is there a way to have tighter git integration with flyte? lets say I want every workflow run to be associated with a git push to a random branch and include a link to the git commit in the ui? the use case is: 1. allowing someone to easily “clone” an entire workflow, edit code associated with some non cached task, and run the dag again 2. viewing the full code associated with a workflow run in a ui somewhere so I can understand exactly what was run; I understand we can see the workflow python file in flyte deck, but not the rest of the code (which I think is saved to s3?)

One way to accomplish that, for example in GitHub, would be to use GitHub Actions. You can create an action that runs on a push, i.e. to a branch, and run the workflow. From the action you can post comments to the PR stream where you can publish the link to the logs or results. To "clone" the workflow would be akin to creating a new branch, editing whatever makes sense in the code (or parameters to the tasks) and pushing that branch to the server. Please note that the workflow will run with the server credentials, so you may need to add some server-side verification. Also, from the push you can trigger other services or webhooks. Other git providers, such as GitLab, have similar mechanisms. Hope this helps.

@curved-whale-1505 can you believe when we first started Flyte was completely tied to git and scientists do not love that

I think I may have confused the requirement when I mentioned git. We can ignore git entirely here, the main ask is to: 1. have a way to view all of the code associated with a task in a ui (pushing to git is one way to do that, but we can probably do something in flyte / flyte deck as well?) 2. have a way for a new scientist to "clone" an existing workflow run, edit some code associated with a task, and run it again I don't necessarily want the scientist to have to push to git and have a github action start the workflow. I was more thinking in line with how flyte already pushes the code to S3 iirc. I want to be able to view that code that's saved to S3 and also allow a scientist to clone it, edit it, and kick off a new run

Great point - yes the code link is buried today - we would love to add it to the template - which is indeed doable

It appears the s3 path is missing for

pytorch

jobs (screenshot 1) but it exists for regular python tasks (screenshot 2)