On Nov 26, 2013 5:44 PM, "Eneko Lacunza" <elacunza@binovo.es> wrote:
>
>
> On 26/11/13 17:21, Ritesh Raj Sarraf wrote:
>>
>> On Monday 25 November 2013 10:29 PM, Eneko Lacunza wrote:
>>>
>>> We have a Debian Wheezy node with the following:
>>> - A directory that is a mountpoint for a SoftRAID (RAID1) device,
>>> with 2 devices. (md0)
>>> * Device one is a local disk (sdb)
>>> * Device two is a multipath network iSCSI disk (dm-3)
>>>
>>> When the server boots up, md0 is degraded because it couldn't find dm-3.
>>>
>> You must have very good reasons as to why you are building an odd
>> configuration like this. But I'm assuming, you do have one.
>
> I know it's odd :) Do you think it wouldn't be so odd without multipath?
>
>>> I have looked to mdraid and multipath init scripts and it seems that
>>> multipath is trying to start multipath devices before networking is up.
>>>
>> Of course, it does. And the reason is that multipath is not a network
>> service. DM Multipath is one of the output targets of the Device Mapper
>> framework, just like LVM2. There are many other consumers of multipath
>> (Fiber Channel etc). Also, there are many people who use ROOT on
>> Multipath, which requires us to start multipath as early as possible, in
>> the boot process.
>>
>> In your case, the problem will not be solved by the _netdev option. The
>> dependency is simple. You need your network first. Then your iscsi
>> devices. Then the MD Raid activation. And then at last you want to call
>> mount.
>>
>> You will have to roll out your own initscript for this requirement.
>
> If the md device had two iSCSI devices, the problem would be the same?
I think not. I believe that, in your case, initramfs is trying to assemble the md device, way before you have the network and iSCSI up. Can you unpack the initramfs and verify it's mdadm.conf?
Ivan