[systemd-devel] What unit file should I depend on?

Manuel Amador rudd-o at rudd-o.com
Wed Feb 1 19:00:53 PST 2012


Thanks, Lennart.  I am aware of the DefaultDependencies=no.

I have posted a pull request to the ZFS on Linux project that does the *early* 
part of ZFS initialization upon boot.  Now the *late* part (initializing pools 
on crypted devices and such) is missing.  This is why I ask these questions:

Pull request including systemd-zfs-generator which I use is here:

  https://github.com/zfsonlinux/zfs/pull/547/files#diff-62

Feel free to post comments or reply with suggestions.

I don't want local-fs.target.wants.  I want local-fs.target.REQUIRES.  Am I 
wrong in wanting that?  Should I want something different?

-------------------------------------------------------------
> Devices show up as they are made available as they are found. There is
> no point in time where "all block devices" have been found. Services
> which assume such a point in time exists are broken.
> 
> Lennart

But there *is* a point in time where "all block devices" required for fstab 
HAVE been found.  That's what I am trying to emulate here.  The only 
difference is that no file systems are registered in fstab -- they have to be 
*discovered* among all the block devices presently connected on the system 
when the system is powered on.

It is rare that one hotplugs a pool to a storage server and expects it to be 
automatically mounted.  Either pools are created in the server using the 
devices already attached to it, or they are imported with an ALTROOT parameter 
such that the imported pool won't clobber existing file systems.  This 
suggests to me that the idea of emulating coldplug behavior by using hotplug 
behavior is not good for ZFS.

All I want is for all pools connected to the system *when it is booted* to be 
properly mounted before local-fs.target finishes.  I thought of splitting the 
process between mounting of filesystems in the root pool and mounting of 
filesystems of *other* pools (such as the ones discovered by fedora-storage-
init.service) to parallelize mounts better.

Am I doing something wrong?  I really would appreciate input on this, as the 
current fstab logic is insufficient for what I am trying to accomplish.  Bear 
in mind that ZFS does not use fstab, hence me creating unit files with a 
generator.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/systemd-devel/attachments/20120201/30b86700/attachment.htm>


More information about the systemd-devel mailing list