I've found a few more things to play around with since the last time I posted in this thread.
1. Use fuse-overlayfs, you need to install the package via apt on the guest first. You can then alter the storage driver in the daemon config. Shut down the CT and change the box in options to enable fuse. I'm not a huge fan of user space mods, yet it does seem to work. Verify data is being written to the new /var/lib/docker/fuse-overfs dirctory. When your dockers are back up, (you saved your compose files, correct?) you can delete the VFS directory in the same location, and you will see a drastic reduction in space. I haven't measured performance, yet it should be more efficient. The problem I've found here is that backups don't seem to work 100% with this configuration. It skips much of the /var/lib/docker/fuse-overlayfs directory. I believe there are ways around this, yet they include shell script and manual invocation of the backups. I prefer to keep in all working in the GUI
2. Create a mount point on the CT that's formatted in ext4 that will hold your docker files. There are instructions for this in the forum, yet I'm having no luck finding the thread. I never got it to work 100% I am unsure as to what functionality you lose. User niziak posted their solution a few posts back in this thread, it seems like a similar solution and is something to try. In fact, I may try it as my next attempt.
3. Stop using docker in LXC's. This is what Proxmox proposes, as they say they don't support docker inside of Linux containers. So, you'd either create a vm (probably the best solution) or move Docker to the host (you'll lose pretty much all the backups/cloning in the UI this way). The downside here is if you're using a shared CPU for LXC's it will no longer work, you need to pass through the CPU to the docker container exclusively.
The exception is if you run a datacenter or workstation card. You can pick up some of the older ones rather cheaply on eBay. Then you'd have to purchase the NVIDIA drivers for virtual GPU, or wait until someone makes an open-source solution. There was a hack available for consumer cards, yet I'm not sure if this still works. It also still requires the purchase of the drivers. You can use the NVIDIA drivers for 90 days for free, yet I haven't encountered a workaround for this time bomb as yet.
So, while I am yet to find the perfect solution this thread seems to get several views, so I'd invite feedback on what other solutions folks have found to this quandary, it should help many who land here via Google.
