grub can't install, server non bootable

[meeshkah]

Ok well let's do a clean install. Maybe I'll be able to cleanly do things now. Messing is learning, in the end

Is there some config reset in Ubuntu LiveCD to reset the config in root? (and hopefully make it working again?)

The last iso I have of Ubuntu Server is 11.10 LTS (x64), will I be able to upgrade to 12.04 LTS only? Is there some config I need to do for the new install to use the raid arrays? If I want to name the lvm I will set on the raid5 binomiale, will I have to rename the one on sdb?

When done, I will edit the first post to reflect the changes.

[darkod]

I would avoid upgrading from 11.10. The point of a clean install is also not tp upgrade right away. I would download 12.04 LTS server and keep it at hand. And I wouldn't try upgrading to 12.10 either, I stick to LTS for servers.

I think the raid will be picked up, but I'm not sure. The partitions with the raid flag will be picked up definitely, so to configure new raid you might need to go into Configure Software Raid while in the partitioning menu.

I'm also not sure if you need to create new raid whether the installer will allow you to have a missing member, like the --create command allows you. You will have to find that out for yourself.

At the end of they day you can always create the raid from all four disks and copy files from your usb backup. In that case i suggest booting the server in live mode first and partition sdb with exactly the same partitions as we did with the other three disks using parted. After that start the server install and create the md0 raid1 of 4 partitions for root, and the md1 raid5 of four partitions for /data.

After you create md1 you can go into Configure LVM and use it as physical device for LVM. If you already deleted sdb you can call the VG the same, the old one will be gone anyway. And don't make the LVs too big for start, only as big as you need them and you barely had few GBs used. The point of LVs is that they can grow as you need them, shrinking is little more complicated. So, start small and grow, not start big and then get into trouble how to shrink them best.

Don't forget in the partitioning step to specify all four swap partitions to be used as swap area, one by one.

That should be all.

[meeshkah]

OK I'll try one last thing. As there semms to be no hope of booting off sdb anymore, I'll repartition it and add it to the raid arrays once configured. Then I will copy the previous root to sdb2 just like the others, install grub to sdb1 and give it a shot. If it still does not boot, then I will do a clean install like you said. I already downloaded the Ubuntu Server 12.04.2 LTS CD.

Questions for you (still):
- if I add sdb (assuming correct partitioning) to md0 and md1, will the mirroring of md0 propagate to sdb2 and populate it with the copy of the root? Or will I have to copy it from the USB? (Am I clear?)
- how do I add sdb2 (assuming correct partitioning) to the raid arrays?

[darkod]

Correct, simply adding it to the array makes it sync. You will see in /proc/mdstat that it will say md0 Syncing until it finishes.

You should add sdb2 and sdb4 with something like:

sudo mdadm --manage /dev/md0 --add /dev/sdb2
sudo mdadm --manage /dev/md1 --add /dev/sdb4

That will do all the work and resync the array onto the new partitions.

[meeshkah]

Oh my, I'm getting now that syncing raid 1 is much more faster that syncing raid 5 since mdadm has to "restripe" all data onto 4 disks instead of 3 for raid5.
So it's a complexity in n squared (n²) = the amount of data to the power of 2. If I'm not mistaken.

It has already synced md0 and is forecasting a 255min-long syncing for data on md1, even though it has no data!
Would it be bad to reboot now or is there no need to wait for mdadm to finish working?

Edit: Damn! I forgot to mkfs.ext4 a partition in sdb2 BEFORE adding it to the md0 RAID array. And the mirroring still occurred.

[darkod]

Edit: Damn! I forgot to mkfs.ext4 a partition in sdb2 BEFORE adding it to the md0 RAID array. And the mirroring still occurred.

That's ok, you don't make the filesystem on the partition. md0 contains it.
I'm not sure if restarting can mess up the syncing. I think it will continue when it boots. The syncing would have been needed in any case when adding the disk, sooner or later.

[meeshkah]

Ok GREAT NEWS! The syncing finished so I could reboot off the Ubuntu Server 12.04.2 LTS amd64 CD. Instead of going with the Ubuntu install I chose for a try to go with the "Boot from first hard disk" option. AND IT DID! The system could not mount /home nor /var/www nor /home/ownCloud as all these were deleted and are on the usb recovery. I'm logged in right now.
In teh end what did the trick was to remake the partitions on sdb from scratch. I'm not sure I understand it fully but it worked.

I think I only have to create the lvm and the lv on the raid5 array in order to mount these volumes. The thing is I don't know how to setup an lvm on md1. mdadm is installed OK and /proc/mdstat returns the arrays OK. The lvm2 package is installed too.

As LVM is already installed, I used the following commands:

$ sudo pvcreate /dev/md1
$ sudo vgcreate binomiale /dev/md1
$ sudo lvcreate --name home --size 200G binomiale # etc for -web and -cloud

(Thanks http://www.howtoforge.com/linux_lvm_p3 !)

I am now wondering what would be best between using ext4 of XFS filesystems? Is there one which could be better for RAID5 use?
The wiki for XFS seems to feature more write optimisation, is that correct?

Your advice?

Edit: And how do I make sure the swap is correctly used?

Edit 2: Based on the conversations and considerations here: http://lwn.net/Articles/476263/ I decided to stick with ext4 fs. Seems much more reliable to me. Quick explanation: bottom line XFS does A LOT more caching than ext4 that may result is A LOT more data loss than ext4. Although that was over a year ago, I fear some of the issued pointed out there are still not solved. So I stick with ext4.

[darkod]

I haven't investigated too much about ext4 vs XFS. If I remember correctly, XFS was more efficient for big files, while ext4 is universal.

So, if you have 85-90% big files, I guess XFS might be better for you. But I haven't used it and and don't know much about it.

Also don't forget you need to do the mkswap on /dev/sdb3 so that it is formatted as swap area and gets an UUID assigned.

I have attached an image of fstab on my home server, you can see how both swap partitions are mounted and with which options. You will need to do that for all four swap partitions.

And after you create and format the LVs you will need to add them to fstab of course (unless you already did that).

Glad you got it going.

[meeshkah]

Ok great news, I'm managed to retrieve my original install without any need for a start from scratch concerning the system. I could get everything back on.

I'm finally back to normal. Ahhh!

Thank you so much for all your help and all your time helping me. Know that you have been my only hope of wiggling my way around a new install for a very long time. I truly thank you for bearing with me that long.

Just one more thing I'd like to do to (hopefully) increase my HDDs life: to make idle hard disks go into standby: http://zackreed.me/articles/60-spin-...dle-hard-disks
I'm not sure that would be very helpful since the root, which is on all disks woudld probably keep all four disks "awake".
May I ask your thoughts on that?

[darkod]

With root on the disks it will be difficult if not impossible to spin them down. For disks that are only for data, should be very easy following the link you posted.
To make them spin down when root is on them would involve palying around and probably disabling many log services, since if various logs write on the disk it will be used and won't spin down.

But later you might regret disabling logging if a problem occurs and you can't troubleshoot it.