[oe] meta-systemd: Introduce a create-volatile to create tmp directories

Peter A. Bigot pab at pabigot.com
Wed Jan 23 20:44:28 UTC 2013


On 01/23/2013 01:50 PM, Holger Hans Peter Freyther wrote:
> On Tue, Jan 22, 2013 at 09:02:12AM -0600, Peter A. Bigot wrote:
>
>> meta-gumstix has this patch and it caused problems with attempts to
>> use systemd_v189 in denzil since it's hidden inside a bbappend to
>> systemd_git.  So I'm hoping to identify a more forward-looking
>> solution.  Is there a plan to fix this issue globally?  Koen
>> preferred using base_files+fs_perms as in Angstrom, but Holger was
>> concerned this would prevent sharing base_files between sysvinit and
>> systemd builds.
> My concern is still there but my approach has some downsides as well. The
> presence of /var/log on tmpfs is fooling the journald to use more memory
> than is healthy. I solved this by disabling the auto-detect in the journald
> configuration.
>
> What is the issue you have with meta-gumstix?

Mostly that I wasn't aware of the patch it added to systemd_git that was 
responsible for creating such a critical part of a functioning Poky 
installation.  I'd had things working with systemd_git in the denzil 
branch as of 21Dec, but when that recipe got reverted around 24Dec I 
switched to systemd_v189 assuming it was compatible.  But things broke 
in a very unexpected way because I lost the meta-gumstix systemd_git 
bbappend that created the volatile hierarchy.

I now understand that it was natural to copy it into systemd because 
sysvinit is responsible for this functionality when it's the init 
manager.  Having to put a bbappend in every variant systemd recipe in 
every distro/BSP/vendor image layer seems cumbersome.  Since my email 
I'm seeing discussion on oe-core on how to integrate systemd in 
Yocto/Poky that touches on managing volatiles, so I'll wait, or 
follow-up there if necessary.

Peter




More information about the Openembedded-devel mailing list