Seeking feedback on GCP VM setup for Coder
Hey all, I’m working on a proof of concept for a dev environment using Coder with ephemeral GCP VMs and would love to gut-check some assumptions and get feedback on a few sticking points.
Setup Overview:
- I'm using Packer to build a custom Ubuntu image with all base OS-level dependencies preinstalled (Docker, Python 3.11, Task, etc.)
- The metadata startup script creates the user and installs the Coder agent
- Devs connect via Cursor
Challenges / Questions:
- Git repo setup: Not sure if I should pre-clone repos in the Packer image (faster boot, but potentially stale), or always clone at runtime using task. What’s worked for you?
- User-level setup: Thinking about using task as the entry point after boot to run nvm install, configure git, clone repos, and spin up containers. Is this a sane division of concerns?
- nvm weirdness: Since it’s user-scoped, I’m running into some complexity around where to install and load it reliably. Any patterns you like?
- Ephemeral VM boot time: Since these VMs are short-lived, I’m trying to minimize cold boot time. Packer helps a lot, but I’m still fine-tuning what should be preinstalled vs. what can be scripted on boot.
Right now I’m trying to make it feel as close to a local dev machine as possible. I'm open to improving the approach with different pieces e.g. devcontainer, installing docker based services directly, etc.
Appreciate any tips or validation, thank you!
Setup Overview:
- I'm using Packer to build a custom Ubuntu image with all base OS-level dependencies preinstalled (Docker, Python 3.11, Task, etc.)
- The metadata startup script creates the user and installs the Coder agent
- Devs connect via Cursor
Challenges / Questions:
- Git repo setup: Not sure if I should pre-clone repos in the Packer image (faster boot, but potentially stale), or always clone at runtime using task. What’s worked for you?
- User-level setup: Thinking about using task as the entry point after boot to run nvm install, configure git, clone repos, and spin up containers. Is this a sane division of concerns?
- nvm weirdness: Since it’s user-scoped, I’m running into some complexity around where to install and load it reliably. Any patterns you like?
- Ephemeral VM boot time: Since these VMs are short-lived, I’m trying to minimize cold boot time. Packer helps a lot, but I’m still fine-tuning what should be preinstalled vs. what can be scripted on boot.
Right now I’m trying to make it feel as close to a local dev machine as possible. I'm open to improving the approach with different pieces e.g. devcontainer, installing docker based services directly, etc.
Appreciate any tips or validation, thank you!
