TIL: Dockercompose Volume Permissions

Snippets of learning.

Error Accessing Mounted USB Drive

I was trying to setup an MLFlow server, with a volume pointing to a mounted external USB HDD:

$ mount | grep DC67
/dev/sdb2 on /mnt/DC67-0ADB type exfat (rw,relatime,uid=1000,gid=1000,fmask=0022,dmask=0022,iocharset=utf8,errors=remount-ro)

With a dockercompose from here, looking specifically at:

services: 
  mlflow-db:
    ...
    volumes:
      - /mnt/DC67-0ADB/mlflow/db:/var/lib/postgresql

Trying to launch the stack (via Portainer), I get the error:

	2025-10-19 00:10:15
	Error	Failure	
Failed to deploy a stack: compose up operation failed: Error response from daemon: error while creating mount source path '/mnt/DC67-0ADB/mlflow/db': chown /mnt/DC67-0ADB/mlflow/db: operation not permitted
	2025-10-19 00:09:04

What’s Happening

Docker can have mount permission issues with external drives (esp. those of exFAT, NTFS, or FAT32). In this case, Docker wants to verify (or change) ownership of the /mnt/DC67-0ADB/* folder. Which, the linux host is saying it not permissible.

Verify Issue

$ mount | grep DC67
/dev/sdb2 on /mnt/DC67-0ADB type exfat (rw,relatime,uid=1000,gid=1000,fmask=0022,dmask=0022,iocharset=utf8,errors=remount-ro)

The drive is mounted with uid/gid=1000 and it is a exfat type.

1. exFAT doesn’t support UNIX permissions (or ownership, or symlinks)
2. Postgres expects to set ownership and permissions in its data directory (/var/lib/postgresql/data… which is a link to the drive)

Solutions

1. Just move the postgres db from the external drive onto the host drive (in this case, 12GB VM of a 60GB Proxmox instance, of a 256GB NVME SSD). This may actually be okay, as the postgres data will likely stay less than 1GB forever. i.e.:

mlflow-db:
  ...
  volumes:
    - /home/b/homelab/mlflow_data/db:/var/lib/postgresql

Which may actually be consisten with the other stack file data that I keep on the nvme drive (e.g.

b@vm-portainer:~/homelab$ ls
caddy_config  caddy_data  heimdall_config  jellyfin_config  plex_config 

2. Reformat the drive to ext4 (linux-native filesystem). This may actually be the right thing to do, as that drive is presumably always going to be holding stuff on this homelab. But that would require a 2TB data transfer/re-transfer and re-mounting.
3. Gamble and run the database with postgres_user, i.e:

mlflow-db:
  image: postgres:latest
  user: "1000:1000"

Which may actually lead to data corruption after reboots.

For today, I’m not a gambler: Option (1)

Still fails: More Permissions

Option 1 will still fail, briefly.

The uid of the postgres default is 999. so we need to (a) make the folder — docker will do this, but it will setup the wrong permissions; (b) change ownership to postgres uid=1000; (c) make sure host (root) can modify:

sudo mkdir -p /home/b/homelab/mlflow_data/db
sudo chown -R 999:999 /home/b/homelab/mlflow_data/db
sudo chmod -R 755 /home/b/homelab/mlflow_data/db

Then relaunch the stack.