You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Given that NFS provides file locking, we could build a lease system around it that can be shared by multiple LiteFS instances. AWS, GCP & Azure all provide hosted NFS mounts so it could be a simple way to run an HA cluster rather than relying on Consul.
The text was updated successfully, but these errors were encountered:
We don't currently have a public roadmap available. Our two main focuses right now are getting LiteFS working better on non-Fly deployments (e.g. AWS, GCP, etc) and to get LiteFS working on ephemeral deployments (e.g. Vercel).
This might be tangential - I can open a separate issue if what I ask about makes sense and it is - but would it be possible to run on fly using the NFS protocol instead of FUSE to provide the custom file system?
For context: I'm asking because Fly is planning to charge more for storing the starting FS state of stopped machines. Some languages can create perfectly OK images with just copying a few files into the scratch base image - but having C dependencies (which might have C dependencies of their own! the terror!) complicates that picture, and FUSE is a C dependency.
I can see that the flyio/litefs docker image is quite minimal - but I don't get the impression it's designed to be fully functional on it's own since the example pulls the runtime C dependencies in as well.
Given that NFS provides file locking, we could build a lease system around it that can be shared by multiple LiteFS instances. AWS, GCP & Azure all provide hosted NFS mounts so it could be a simple way to run an HA cluster rather than relying on Consul.
The text was updated successfully, but these errors were encountered: