Rclone dir cache remains when mount drops

Help and Support
1

I have a rather unique issue and I'm not sure if it's intended behavior or not.

I have a mount with static, read only content that doesn't change. To improve performance, I have rclone mount this to a new mount point with a high dir cache value.

Rclone version 1.70.2 (has happened with last 3 or so versions as well.)

rclone mount /mnt/src /mnt/dst
--uid=1000 --gid=1000 --umask=002
--allow-other
--read-only
--timeout=1h
--dir-cache-time=120h
--attr-timeout 120h
--vfs-cache-max-age=4h
--vfs-cache-min-free-space=80G
--vfs-read-chunk-size-limit=128M
--vfs-read-chunk-size=8M
--vfs-fast-fingerprint
--vfs-cache-mode=full
--links
--cache-db-purge

This works great. Once in a blue moon, if I need to restart the mount, or some network connectivity issue happens which causes the mount to drop and for whatever reason and not remount properly, so the doing an ls /mnt/dst results in Transport endpoint not connected.

To try and automate remediation of this, I have a script that fires every minute that looks for /mnt/dst/mounted.flag. if this file doesn't exist it forces a fusermount -uz /mnt/dst then remounts. To clarify this is when the rclone dst drops, not the source.

However, to my surprise, doing an ls /mnt/dst/mounted.flag returns the file listing even when an ls of the root, /mnt/dst returns Transport endpoint not connected. I'm guessing this is the dir cache staying alive?

I'm not sure how to reproduce this consistently, by that I mean make the mount fail while the rclone process is still running, otherwise I would provide those steps. I do know if I have rclone set as a systemd service, with the execstop doing fusermount -uz /mnt/dst, if I restart the service sometimes it restarts, sometimes it yields this failed state.

So I guess my questions are two told:

  1. Is it still returning the file because of dir-cache and if so is there a better way to do a health check against the mount?
  2. Is the systemd service restart failure a common/known issue and any known remedies?

Thanks in advance.

2

can you post the service file?

afiak, based on what you posted, does nothing, can be removed?

can you post a rclone debug log?

3

Thank you for the reply. Based on this post:

I thought that flag would clear any vfs cache on the system on start of the mount. Is this not the case?

Wrt to debug log I can try to capture but as I noted it somewhat random and difficult to reproduce.

Systemd service:

Description=backup Daemon
After=multi-user.target
[Service]
Type=notify
ExecStart=/usr/bin/rclone mount /mnt/src /mnt/dst \
--uid=1000 --gid=1000 --umask=002 \
--allow-other \
--read-only \
--timeout=1h \
--dir-cache-time=120h \
--attr-timeout 120h \
--vfs-cache-max-age=4h \
--vfs-cache-min-free-space=80G \
--vfs-read-chunk-size-limit=128M \
--vfs-read-chunk-size=8M \
--vfs-fast-fingerprint \
--vfs-cache-mode=full \
--links \
--cache-db-purge
ExecStop=/bin/fusermount -uz /mnt/dst > /dev/null
TimeoutSec=120
Restart=always
KillMode=process
User=0
Group=0
[Install]
WantedBy=multi-user.target```
4

Yes but not for VFS cache you are using. It is an old flag for deprecated cache overlay remote.

5

the only way to clear files in the vfs file cache is to delete them using rm

well, there are two vfs caches, check out my summary of vfs caches

6

Got it. Thank you. But is it expected behavior for dir cache to remain and respond when the mount drops?

7

no. if you can show different, post the details...

8

Will do. What would be the best way to illustrate this, mount with debug logging to a file, then try to reproduce?

9

not sure what you mean by source and dest?

afiak, with rclone mount, there is not really a concept of source and dest.
rclone mounts a remote as a local directory.

not sure what you mean by drop?
as long as rclone executable is running, so too, are the caches.