Adds mounts to fly backend

[benbot]

I'm working on getting some elastic LLM inference set up using phoenix, fly, and FLAME and needed volumes attached to my flame workers so that I don't need to download an 8gb+ model on startup.

I added support for the mounts field in the machines api (https://fly.io/docs/machines/working-with-machines/)

Though it seems like fly has a bug where this isn't working right now. I keep getting 500s when passing in that argument. I made a thread on their forum to track it
https://community.fly.io/t/machine-api-returning-500-when-passing-in-mounts/17336

[benbot]

Quick update. Apparently the volumeId is a required field, so i'm going to work on a change to this to grab a volumeId from existing un-attached volumes with matching names, and use those ids to build the mounts array

[benbot]

Alright, now the fly backend will attempt to find a volumeId that matches the name specified in the mounts config option.

If it finds enough it'll add those ids in as volume in the fly machine request.

This gets around the 500 that's thrown by fly if you don't include the volume id.

[mruoss]

Stumbled upon this. Some suggestions to improve readability.

[benbot]

These are great suggestions!

I'll get them in tomorrow

[andr-ec]

@benbot could this be updated to resolve conflicts? I'd love to see this merged in to the latest!

[pxp9]

hey, you should be aware that FLAME is not using Jason module for last versions it is using FLAME.Parser.JSON which is a wrapper which detects the version you are using and sets Jason or :json parsers.

[pxp9]

Hey @andr-ec , I have opened a new PR which continues the work from Ben

Hope it is what you were looking for.

[benbot]

I check back in on flame every once in a while and it's kind of crazy that there's been 3 PRs for this same feature over the years and none of them got any official feedback

I know this isn't a priority for phoenix or fly, but still... total silence :(

I'll close this one, since the others are more up to date anyway.