Re: [OSR-devel] GlusterFS and mdadm - fsck patch

[Gordan B. <go...@bo...>]

Marc Grimme wrote:

>> I'll have a look, but wouldn't glfs be a little different? It's a
>> 2-step mount, first the underlying backing fs (in my case ext3,
>> but it could be anything), and then glfs on top of that. At a
>> glance, the "local device=$1" would return the glfs volume spec
>> file, not the underlying
>>
>> backing fs which is what would actually need to be fsck-ed.
>>
>> That means that at the very least what is passed to 
>> glusterfs_fsck_needed would need to be different. It would need to be
>>
>> rootsource, rather than rootvolume.name.
 >
> Ok. Got it. I didn't see that (forgot about the glusterfs dependencies and fs being mounted in clusterfs_services_start).
> So there is no other way. I cannot move the rootfsck functionality before clusterfs_services_start cause there are clusterfilesystems that require a running cluster before being able to fsck the rootfs and it belongs at this place.

Yes, I understand. glusterfs operates on a level above the usual one.

> I don't "like" your approach cause fsck does not belong in glusterfs_services_start but I don't see any other way.

Technically, I'd argue it does belong there, because the underlying 
backing fs is in fact a "service" (or maybe we should call it a 
"resource", but let's not nitpick) that glusterfs depends on.

> Give me one or two days to think about and decide if there are other ways.
> I'll let you know in a few days. But I think we'll do it the way you have proposed.

OK, let me know what you decide.

Gordan