On Tue, Aug 23, 2011 at 17:06, Hannes Reinecke <hare@xxxxxxx> wrote: > On 08/22/2011 06:35 PM, Kay Sievers wrote: >> >> On Mon, Aug 22, 2011 at 17:45, Hannes Reinecke<hare@xxxxxxx> wrote: >>> >>> On 08/22/2011 03:47 PM, Christophe Varoqui wrote: >>>> >>>> On jeu., 2011-06-30 at 14:10 +0000, "Jóhann B. Guðmundsson" wrote: >>>>> >>>>> Is anyone taking a look at porting the multipathd legacy sysv init >>>>> script to native systemd unit file? >>>>> >>>>> I filed 690828 on 2011-03-25 and yet to receive somekind of response on >>>>> it and unless some action is taken then a) device-mapper-multipath will >>>>> be blocking the Fedora 16 alpha release or b) be ripped out of @base >>>>> and >>>>> be made sure it does not get installed on the official livecd. >>>>> >>>> What is the status of this request ? Has anyone stepped-in to the job ? >>>> If not I'll look into it. If you have Fedora-specific systemd >>>> implementation guideline, please point me to the appropriate >>>> documentation. >>>> >>> I have talked with Kay& Lennart about this. >>> >>> The main problem here is that we don't / can't know if a device is >>> multipathed or not; this is basically a policy decision. >>> Even a simple switch (multipath on/off) wouldn't suffice here, as this >>> particular device might be excluded via the configuration file. >>> >>> Idea here is to modify the 'multipath' program to determine if a >>> particular >>> device is handled by multipath or not. If so, _no other_ configuration >>> must >>> be done in that device. >>> The multipathd daemon itself already listens on events from the udev >>> socket; >>> it should be pretty trivial to convert that to socket activation. >>> >>> And, of course, we need to solve the mount problem. Last I've heard was >>> that >>> systemd _won't_ be handling mounting, which in effect means we cannot >>> mount >>> any systems provided via multipath ... >>> >>> What is the status here? >> >> It fully support mounts on boot, dynamically created, or on hotplug, >> with or without automounter support. >> >> But it will not assemble/handle/know anything about any dm/md devices. >> > Yeah, but this should taken care of if and when LVM2 and MD gets converted > to systemd, right ? Right. The tool to do that would probably be some raid-assembly daemon that replaces the insanity in current shell scripts and called tools that scan /dev. There is no plan currently to integrate any raid-assembly into systemd natively. Kay -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
