Host Startup; Booting into LVM Disk Fails

[dholtz-docbox]

I am looking to have someone open the enclosure so I can swap the cables - darn special screwdriver that I don’t have. Until then, I am looking at how to handle how /dev/sda is assigned, and it looks like it comes down to udev rules? That’s what I am looking at right now. I know a lot of this is hitting a bridge I was looking to prolong, but I have been looking to more intimately setup the partitions. I guess this is the start of understanding how some of this is handled.

[george1421]

@dholtz-docbox Remember at this point in time you are NOT in any OS. When the iPXE menu exits you are still running the iPXE kernel. It knows nothing about your target OS. Its just passing (or trying to) pass off boot control some some boot device.

[Wayne Workman]

George is right on that.

[george1421]

@george1421 OK I have a hackish way to test this idea.

But first, I went through each of the exit modes and all pick the first hard drive. Looking into the code all of the exit modes are hard coded to boot to the first hard disk.

Sooo… as a test please edit this file /var/www/html/fog/lib/fog/bootmenu.class.php search for some fragment of “sanboot --no-describe --drive 0x80” it should be listed only once in that file. Change the 0x80 to 0x81 then file and save the class file. Be sure that you have the host set to sanboot and then use the browser trick I chatted to you to confirm that the sanboot is now sending to the 0x81 disk. Once confirmed try to boot that target computer again. If it works then we have to find the next steps, but we can prove the process works.

[dholtz-docbox]

That makes sense. I guess my thought here, is, before trying to boot through network, after installing Ubuntu, I run a set of udev rules, this remapping which disk /dev/sda and /dev/sdb are associated with.

Also, I can’t switch the cables around, one drive is SATA where the other is… like an onboard PCI-e disk. I wanted to handle it when installing Linux, but it looks like they have their own set of udev rules which always maps the larger drive to /dev/sda instead /dev/sdb. Which is why I was thinking, maybe we’ll just have our own we run as necessary to correct this, prior to building our golden image.

At the end of the day, the primary disk should be /dev/sda anyway, so I want to figure out how to get the smaller drive mapped to /dev/sda instead of /dev/sdb if I can.

Anyway, my time has run up for today. I will try a few more things tomorrow morning, specifically regarding udev rules first, most likely.

Thanks again for everyone’s help.

-Dustin

[dholtz-docbox]

@george1421 said in Host Startup; Booting into LVM Disk Fails:

@george1421 OK I have a hackish way to test this idea.

But first, I went through each of the exit modes and all pick the first hard drive. Looking into the code all of the exit modes are hard coded to boot to the first hard disk.

Sooo… as a test please edit this file /var/www/html/fog/lib/fog/bootmenu.class.php search for some fragment of “sanboot --no-describe --drive 0x80” it should be listed only once in that file. Change the 0x80 to 0x81 then file and save the class file. Be sure that you have the host set to sanboot and then use the browser trick I chatted to you to confirm that the sanboot is now sending to the 0x81 disk. Once confirmed try to boot that target computer again. If it works then we have to find the next steps, but we can prove the process works.

Haha, yeah… I tried that earlier. The disk just boot looped with no prompt, indefinitely, until I force-killed-power to the machine.

Edit> That was my immediate hack idea too, just to see if it would work
PS> Everything checked out, regarding your hack for reviewing the settings passed - 0x81 was assigned when tested.

[Wayne Workman]

@george1421 said in Host Startup; Booting into LVM Disk Fails:

Sooo… as a test please edit this file /var/www/html/fog/lib/fog/bootmenu.class.php

Along the same lines, you will find this at line 142 concerning grub that is easy to snipe:

'rootnoverify (hd0);chainloader +1`

Maybe change hd0 to hd1 and switch the host to grub and see what happens.

[dholtz-docbox]

@Wayne-Workman : No dice - good suggest. It just takes me to the GRUB4DOS terminal.

I feel at an impasse, as I know I can leverage FOG as the solution I am looking for. However, I don’t think we have figured out how to structure our partitions so that it works well with remote imaging. If that is the case, I think it will take us a little longer to finalize everything, but I will start looking into how to install Ubuntu the way we need it - where the OS disk is /dev/sda, and the data disk is /dev/sdb. It feels kind of silly, but I feel like a lot of issues would be cleared up by figuring that out first. I just wish the Ubuntu installer let me map them to sda or sdb, it always assigns them automatically. I wish I had more control over it, I guess.

-Dustin

[george1421]

@dholtz-docbox Can you do what I asked (I think, or at least I was thinking about it). Schedule a debug deploy to that target system. Then at the FOG Engine command prompt, key in lsblk and post the results here.

Second, when you installed ubuntu you are sure that you didn’t install the any type of boot loader onto that first drive?

Third, In the AMD bois for this, when you tell it to boot from the disk with the ubuntu on it. In bios is this listed as the second hard drive? So the bios is seeing it as disk 2 or is it seeing it as another number disk? (I’m not sure if number is base 0 or base 1 for this discussion).

I still plan on testing a similar setup with a dell 780 and 2 hard drives to see if I can boot off the second drive.

[dholtz-docbox]

@george1421 : Let me get started on the first, and I will return with the rest.

Edit> Output from debug…

[george1421]

@dholtz-docbox Well thats a bit unexpected.

sda is a unpartitioned ~1tb disk

sdb is a bit confusing since there are two small partitions and then a large partition (where the lvm probably is). The two small partitions is what I’m questioning. One I might understand since that is a boot partition since you are using lvm on sdb3.

sdc looks like maybe a 8GB flash drive?

Is this an accurate assessment?

[dholtz-docbox]

@george1421 said in Host Startup; Booting into LVM Disk Fails:

Second, when you installed ubuntu you are sure that you didn’t install the any type of boot loader onto that first drive?

When I install Ubuntu, I go through the guided installation of LVM, using the entire disk. From there I let it create and write the partitions. I only install Ubuntu on the second disk.

I haven’t done anything to the other disk yet - it’s still just a factory-default drive. I don’t have an option to format this disk when I go through the process, at least I don’t feel like I do. I am use to the desktop installer, I guess. That said, should I try to also format this drive?

@george1421 said in Host Startup; Booting into LVM Disk Fails:

Third, In the AMD bois for this, when you tell it to boot from the disk with the ubuntu on it. In bios is this listed as the second hard drive? So the bios is seeing it as disk 2 or is it seeing it as another number disk? (I’m not sure if number is base 0 or base 1 for this discussion).

I posted this earlier in the thread, let me dig it up. That said, I am pretty sure it’s…

1. NIC
2. ubuntu (not sure who this one is, but it’s the only one I can boot from)
3. ubuntu (not sure who these guys are)
4. sata drive (which should be /dev/sdb)

@george1421 said in Host Startup; Booting into LVM Disk Fails:

I still plan on testing a similar setup with a dell 780 and 2 hard drives to see if I can boot off the second drive.

Oh, that’s awesome! Thank you very much.

[dholtz-docbox]

@george1421 said in Host Startup; Booting into LVM Disk Fails:

@dholtz-docbox Well thats a bit unexpected.

sda is a unpartitioned ~1tb disk

sdb is a bit confusing since there are two small partitions and then a large partition (where the lvm probably is). The two small partitions is what I’m questioning. One I might understand since that is a boot partition since you are using lvm on sdb3.

sdc looks like maybe a 8GB flash drive?

Is this an accurate assessment?

Pretty spot-on

Let me take a snapshot of what the partitioner does to the disk…

Edit> Installer Album
Edit> This is how it looks after installing ubuntu, all I do is move the IBA option to the top for network boot priority

[george1421]

@dholtz-docbox can you move the values at option 3 and option 4 into the slot 1 and 2 positions?

With a quick registration and then a boot to to iPXE and then local hard drive booted the ubuntu boot loader correctly with the default sanboot.

[Wayne Workman]

This is just a wild idea I have that probably will not work but what if you set the primary hdd in fog to sdb for capturing and set another hosts primary disk to sda and deploy ? Would the deployed machine pass through the hdd exit process fog has?

[dholtz-docbox]

@george1421 : I feel I have tried this, but let me give it a go. I have tried so many permutations of these settings, it’s hard to say which combinations haven’t been checked, heh. Do you have the Host Primary Disk setup too?

[george1421]

@Wayne-Workman After I finish a project I’m working on I’ll document what I did, and it worked correctly. The system always booted ubuntu from a second disk even when I swapped the initial /dev/sda out with a virgin hard drive.

[george1421]

@dholtz-docbox said in Host Startup; Booting into LVM Disk Fails:

@george1421 : I feel I have tried this, but let me give it a go. I have tried so many permutations of these settings, it’s hard to say which combinations haven’t been checked, heh. Do you have the Host Primary Disk setup too?

I did nothing more than install a second hard drive in the 780, the original first hard drive had windows on it, installed ubuntu on the second hard drive changed the boot order for second hard drive first then nic booted and ubuntu booted, then swapped the nic and ubuntu disk, pxe booted, registered using quick registration, rebooted into the ipxe menu and then let it timeout to boot the hard drive. It booted ubuntu, from there (and just to make sure it didn’t install a boot loader on disk 1) I swapped disk 1 out with a virgin ssd, pxe booted, let the timeout happen and then it booted into ubuntu. I never touched the fog web gui, my default bios exit mode is sanboot in FOG.

[dholtz-docbox]

@george1421 : Hmm… There must be something on /dev/sda because quick registration handled it for /dev/sda instead of /dev/sdb. That said, when it boots, it boots with an empty cursor. It would state it in lsblk if it had a boot part, right? I feel we are close…

[george1421]

I wanted to update the thread on where we currently are with this issue.

I had a several hour chat session with the OP. What it appears is this device they are building is uefi based. It appears the disk appears to also be setup as GPT. What is really confusing me right now is the OP said this device boots with undionly.kpxe and ipxe.efi being sent to the device. This is the first I’ve ever see a device be dynamic like this? But if the device is in UEFI mode that would explain the firmware configuration boot order, why ubuntu is specifically called out as a boot device. I asked the OP to contact the device manufacturer to find out for sure what mode/format this device is. When I say device, this is not a typical computer but an embedded device in some kind of equipment that runs stock ubuntu (14.04 I think). So if this is uefi then playing with the bios exit modes will not have any impact on the boot quality since the uefi exit mode setting in fog would be in play. Setting the exit mode for uefi to refind did not help. My personal opinion is that there is a combination of issues here including the way the image is created on the destination disk. (this here is a personal opinion). The way it is currently setup the OP has 2 disks installed in this device (that is OK), and they are deploying ubuntu to the second disk (still ok). But from there the OP created a single lvm partition and installed the OS and swap on that single lvm. This in its self is not bad, but from my personal opinion I would just create the disks using traditional partitions and leave the LVM overhead out of the picture since the unbuntu disk space will never grow where you might need the LVM features. There is no right or wrong way here, I would just use normal partitions to make it easier to manage.

I left the OP with the task of setting up a similar setup using a VM to prove out if their design and fog will be successful. I recommended that they use a vm to create their golden image anyway, that is how we create them in the windows world. The VM will also abstract the golden image from the underlying hardware, plus it will give them the ability to create snapshots and roll them back as they perfect their golden image. Once they have it perfected then capture it with FOG and deploy to the target hardware (once they fully understand how that device is defined).

At this point FOG is working as its was designed, there are just some unknowns with the hardware that need to be worked out before he can start imaging these devices.