Rclone calls Cache.reload on startup, to check if any files from the VFS cache have to be uploaded. Since this talks to the remote, it can take a long time. After the default --daemon-wait setting of one minute, rclone exits and mounting fails. Increasing --daemon-wait leads to a gRPC timeout after 2 minutes, also causing Kubernetes to see the mount as failed, while the rclone process still exists and now never exits, because NodeUnpublishVolume is never called.
Not running with --daemon and managing the child process manually, as seen here might be a good idea (though it requires checking that the mount was successful), but can't really solve this problem, because we still have to finish mounting before the 2 minute gRPC timeout hits us.
It seems to me that the only solution is to start with a fresh cache each time, which brings its own problems, i.e. data loss.
Rclone calls
Cache.reloadon startup, to check if any files from the VFS cache have to be uploaded. Since this talks to the remote, it can take a long time. After the default--daemon-waitsetting of one minute, rclone exits and mounting fails. Increasing--daemon-waitleads to a gRPC timeout after 2 minutes, also causing Kubernetes to see the mount as failed, while thercloneprocess still exists and now never exits, becauseNodeUnpublishVolumeis never called.Not running with
--daemonand managing the child process manually, as seen here might be a good idea (though it requires checking that the mount was successful), but can't really solve this problem, because we still have to finish mounting before the 2 minute gRPC timeout hits us.It seems to me that the only solution is to start with a fresh cache each time, which brings its own problems, i.e. data loss.