i want to know, when i deploy in argo-workflow, why metaflow was package in code instead of pip install metaflow ?

three main reasons: 1. Imagine having a 100-way foreach. It’d need to

pip install metaflow

in 100 tasks in parallel, which often leads to failures. 2. Sometimes Metaflow is used in environments where you can’t pip-install packages freely. 3. We want to guarantee that the same Metaflow version is used in all tasks, no matter what’s in pypi. Note that you can use a local version of Metaflow that’s not in pypi.