Few things prove what a fucking horrible language Python actual is than the work required to get the NVIDIA PyTorch container image to run in JupyterHub in Kubernetes. The Dockerfile work is fucking insane because of all the requirements to first install Anaconda, then Mamba, then whatever all the dependencies are.
Imagine for a minute if all this had been done in Go. There would be one, maybe two compiled binaries with all their dependencies already resolved, no need to fuck around with grabbing multiple tools just to setup the framework to pull down more dependencies just to have to fight with where they all end up and the user permissions involved.
People can have their shallow opinions for Python without any actual work using the language at scale. That’s their right. But I have yet another pile of objective steaming proof that Python is shit. Unfortunately, JupyterHub is entirely Python and started as a way to get Python on the Web. And not I am stuck holding the fucking pile of shit to get it running in K8S. No wonder no one on planet Earth has gotten this to work so far (according to multiple Internet searches).
On the bright side, this has given me yet another idea that I could make into a billion-dollar startup if I wanted to. JupyterHub needs to be replaced with a JupyterHub compatible turn key solution that is designed from the ground up for the modern Kubernetes/cloud-native era. It could still support all the JupyterHub stuff for compatibility, but could even usher in the new age of Go as the ML replacement for Python, which a few very selective ML experts have already adopted (which surprised me). These people realize that the future of large, clunky, disastrously bad scripting languages that cannot be cross-compiled and easily deployed is fucking over.
Tags:
#rants #python #jupyter #cloud #k8s #golang