Hey there,
I have some trouble with the docker_data.vhdx growing increasingly large and not shrinking after the process is done, on a win11 machine.
I’ve already tried the optimize-vhd method described here, but the gains are very small, to not effective at all.
So I would like to better understand what is happening.
Currently desktop_data.vhdx sit at 146.8GB (but it ballooned to eat all the available space in previous run) My setup is the following:
- installed
webodmfollowing the guide on the github - installed a
nodeodm:gpuinstance following using the command in the github
Then I’ve run a couple of process (around 700 pictures) using the gpu node, with the following settings (notice optimize-disk-space:true):
auto-boundary:true, auto-boundary-distance:20, dsm:true, dtm:true, feature-quality:high, mesh-size:300000, optimize-disk-space:true, pc-quality:high
In one job I had the image resized to 2048px, in the other I kept the original (~3.6GB of photo). The two completed job claims to use the disk:
- 4.88 GB (full photos)
- 3.15 GB (resized photos)
The diagnostic information says 39.6GB are being in use. I would expect to be able to resize the docker_desktop.vhdx to around that size, instead of the current 146.8GB, but using the optimize-vhd command yield no result at all. By experiece I know that if I run another job, the size of the vhdx will continue to grow.
running docker system df I get around 40GB of used disk:
TYPE TOTAL ACTIVE SIZE RECLAIMABLE
Images 5 5 10.98GB 80.03MB (0%)
Containers 6 6 9.922GB 0B (0%)
Local Volumes 3 3 19.7GB 0B (0%)
Build Cache 0 0 0B 0B
So I don’t really understand why the disk size of the docker_desktop.vhdx is not shrinking to a more reasonable size?
I tried to delete the jobs from the webodm interface, and diagnostic page says around 23GB of disk used afterwards, but running optimize-vhd nothing change.
The only way to meaningfully reduce the size of the vhdx is to delete the nodeodm:gpu container. Doing this, the size is reduced to 25.1GB, without event needing to run the optimze-vhd command (I just exited docker-desktop and checked the size).
I guess I can spun new nodeodm:gpu containers at will and destroy them afterwards, but doesn’t feel the right way? Also after a job shouldn’t the nodeodm:gpu be “empty”, all data passed back to the webodm container?
(NOTE: I’ve set sparseVhd=true in the .wslconfig file)
