Hey folks, hoping someone can shed some light on a recurring headache in my homelab setup.

I'm running NanoClaw agents in a Docker Swarm (compose v3) to handle automated backup tasks for a few small services. The tasks are defined to run on a schedule, each targeting a specific app's volume. Lately, I've been seeing random failures where a task logs `'device or resource busy'` and quits. It doesn't happen every time, and it seems more frequent when scheduled tasks overlap, even if they're for *different* services.

My understanding is that NanoClaw's container model should isolate these tasks completely. They each get their own temporary container, spun up from the agent image. The confusion is around the shared volumes—while the tasks themselves are isolated, they're all mounting the same host bind-mount (`/mnt/docker-volumes`) to access the data they need to back up.

My suspicion is that the isolation breaks down at the host volume layer, especially if:
* Two tasks try to read the same source volume (even for different apps) at exactly the same time.
* The underlying filesystem (ext4) or a Docker driver has a lock on a file/directory.
* There's some cleanup lag from a previous container holding a reference.

Has anyone else hit this? I'm trying to build a reliable, compliant audit trail of these backups, and random failures are a real problem for my policy. I'm looking for practical fixes—would staggering schedules be the only cure, or is there a way to configure the volume mounts or task isolation to prevent this?

--Emily