[systemd-devel] journal on btrfs subvolume, failed unmounting /var/log/journal

[Chris Murphy]

On Feb 14, 2014, at 4:19 AM, Lennart Poettering <lennart at poettering.net> wrote:

> On Tue, 11.02.14 21:02, Chris Murphy (lists at colorremedies.com) wrote:
> 
>> It looks like var-log-journal.mount is called before systemd-journald
>> is finished using it. Is the failure message merely cosmetic? The file
>> system comes up clean on reboot so I'm pretty sure it's being cleanly
>> unmounted despite this message.
> 
> Yeah, there's a bit of an ugliness here: journald stays running until
> the final killing spree which keeps the fs busy. The main reason for
> that is simply that we have no nice way to tell journald synchronously
> that it should stop access to /var and continue logging into
> /run. Normally we'd use dbus for requesting this, but since dbus-daemon
> is a client of journald and we really can't have the cyclic dep betwen
> journald and dbus-daemon we cannot make journald be on the bus. With the
> advent of kdbus this will change however, as then we have dbus around
> all the time, and the cyclic dep is gone. 
> 
> The issue is mostly cosmetic as we still will cleanly unmount all
> remaining file systems right before shutting down. We will also cleanly
> terminate journald then. 

OK no problem.

I expect the same issue, but I'm thinking it's better to put /var/log on its own subvolume instead of /var/log/journal. Then it enabled immunity from rollbacks for other logs, in addition to the journal. This way all logs can be kept persistent regardless of what subvolume the user boots from. And this is also applicable for LVM Thin Provisioning layouts.

Chris Murphy