I created a volume of 2GB and I resized it later to 3GB, I remember checking using df after the resize and it was fine. A few days later (and a few deploys too) when the app was downloading something that should fit within the 3GB I get an error that I’m out of disk space. I ssh into the instance and the volume is now 2GB although if I do fly volumes list is shows up as 3GB.
I destroyed it (you can actually do that while it’s attached to an instance - I don’t think you should be able to) and created a new 3GB one and restarted my app. That worked fine.
In conclusion, do not rely on the volume extend feature (flyctl volumes extend).
If the fly team wants to investigate further, the volume was vol_pkl7vzkjjk9vqg60.
I suspect this will be hard to investigate now as the volume was deleted.
If this was a V2 app then is it possible the app was not restarted after extending the volume? This step must be manually performed for the app to pick up the new volume config.
This may explain the discrepancy between the size of the volume when ssh into the instance vs. fly volumes list.
It was a nomad app (v1) and if I remember well it automatically got restarted or I restarted it since I cheked the size and it was 3GB. I suspect that the original value was cached somewhere or the original 2GB volume was still available somewhere since it only went back to 2GB later (not sure when).