Fuse manager restarts if stargz restarts, causing the remote snapshots to remount

[dgaponcic]

Based on the docs on fuse-manager (https://github.com/containerd/stargz-snapshotter/blob/v0.18.2/docs/overview.md#fuse-manager-overview), we can use a fuse daemon to manage the mounts independently from the stargz process.

I used the documentation, and enabled the fuse-manager by adding the following stargz conf:

[fuse_manager]
enable = true
# address must be an absolute path; default is "/run/containerd-stargz-grpc/fuse-manager.sock"
address = "/run/containerd-stargz-grpc/fuse-manager.sock"
# set a custom binary path if the executable is not in PATH
path = "/usr/local/bin/stargz-fuse-manager"

When I start stargz, I see the /run/containerd-stargz-grpc/fuse-manager.sock socket, and the fuse-manager process:

 354206       1 /var/usrlocal/bin/stargz-fuse-manager -address /run/containerd-stargz-grpc/fuse-manager.sock -fusestore-path /var/lib/containerd-stargz-grpc/fusestore.db -log-level info

Also I see the db file, and it gets populated.

The problem

When I stop stargz, the mounts are being preserved, which is perfect. But when I start stargz, all my mounts are being remounted, which breaks the logic in my containers.

You can see here that i have the fuse-manager running, but when I stop stargz the process is gone, and when I start it again a new fuse-manager process is started (which explains the remounts).

$ date; pgrep -f stargz-fuse-manager
Tue Feb 10 15:32:03 UTC 2026
357797
$ systemctl stop stargz-snapshotter.service 
$ date; pgrep -f stargz-fuse-manager
Tue Feb 10 15:32:09 UTC 2026
$ systemctl start stargz-snapshotter.service 
$ date; pgrep -f stargz-fuse-manager
Tue Feb 10 15:32:26 UTC 2026
358416

Am I doing something wrong? Can I de-couple the fuse-manager and stargz completely? So that one can be restarted without triggering remounts?

Best regards,
Diana

[ktock]

@dgaponcic The fuse manger is spawned by containerd-stargz-grpc and still belongs to the same service so it is killed when you stop the service. You can have a separated service for the fuse manager to prevent being killed by systemctl stop stargz-snapshotter. It can be something like the following:

[Unit]
Description=stargz snapshotter FUSE manager
After=network.target
Before=stargz-snapshotter.service

[Service]
Environment=HOME=/root
ExecStart=/usr/local/bin/stargz-fuse-manager --address /run/containerd-stargz-grpc/fuse-manager.sock --fusestore-path /var/lib/containerd-stargz-grpc/fusestore.db --log-level=debug
ExecStartPost=/bin/sh -euc 'until [ -S /run/containerd-stargz-grpc/fuse-manager.sock ]; do sleep 0.1; done'
Restart=always
RestartSec=1

[Install]
WantedBy=multi-user.target

When containerd-stargz-grpc finds the fuse manager socket on startup, it just uses that socket and doesn't start a new fuse manager process.