[systemd-devel] machinectl remove always fails with "Device or resource busy"

[Lennart Poettering]

On Sat, 30.04.16 23:59, Benoît Zugmeyer (benoit at zugmeyer.com) wrote:

> Hi list,
> 
> I'm experimenting with machinectl/nspawn to deploy multiple webservices on
> a single host. I have some trouble to remove unused images. Here is what
> I'm trying to do:
> 
> # machinectl terminate preprod
> 
> # machinectl list
> MACHINE CLASS SERVICE
> 
> 0 machines listed.
> 
> # machinectl list-images
> NAME     TYPE      RO  USAGE  CREATED                      MODIFIED
> preprod  subvolume no  516.9M Sat 2016-04-30 21:30:06 CEST n/a
> preprod2 subvolume no  470.5M Sat 2016-04-30 22:33:53 CEST n/a
> preprod3 subvolume no  486.6M Sat 2016-04-30 23:04:22 CEST n/a
> 
> 3 images listed.
> 
> # machinectl remove preprod
> Could not remove image: Device or resource busy
> 
> # systemctl restart /var/lib/machines
> 
> # machinectl remove preprod
> Could not remove image: Device or resource busy
> 
> I've got this error even after (physically) rebooting the host.
> 
> I tried searching the Web to find some resources, but the only related
> result I can find is a bug report about disk quota[1], but I don't have any
> issue about disk quota.
> 
> Is it a known issue? is there any tool to help me diagnose this? I tried
> toying around with lsof and fuser but without any success, but I am not an
> expert.

No this issue has not been reported yet.

Could you please file an issue on systemd github, and include the
output "strace -f -s500 -p $PID -o log" generates in the file called
"log" while you issue a single "remove" call like this? Replace $PID
by the PID of the machined process. Invoke this strace command
immediately before you run the machinectl remove command, from a
different window, and then terminate the strace invocation with
Ctrl-C after the machinectl command failed. Then attach the contents
of said "log" file, it should tell us why the image is considered busy.

Lennart

-- 
Lennart Poettering, Red Hat