raid 1 goes across all the drives / disk management in UI not working correctly

[Aleksandr Shaposhnikov]

Raid 1 got created across all the drives in the system.

If system has 40 drives all of them will be used for boot partition.
This isn't good behavior. Fuel should allow to select how many drives should be used for such thing. One/Two/All/Etc.

[Aleksandr Shaposhnikov (alashai8)]

Add:

Even if only one drive configured in Web UI all available drives will be used.

[Aleksandr Shaposhnikov (alashai8)]

Workaround:

Modify the following file and restart supervisor right after the mirantis-openstack masternode installation or before creating a cluster.

/opt/nailgun/lib/python2.6/site-package/nailgun/volumes/manager.py:557

-boot_is_raid = True if disks_count > 1 else False
+boot_is_raid = False

/etc/init.d/supervisord restart

[Aleksandr Shaposhnikov (alashai8)]

Workaround not working. It only disable raid but disk space still used for /boot physical volumes and all the stuff. Basically this mean that disk management system in UI not working at all.
This is critical bug.

[Mike Scherbakov (mihgen)]

> This isn't good behavior.
why? What is the user impact?

Roman, why do you raise the priority to High, too?

We have boot partition on every HDD for the following reason - we can't get BIOS config - which disk to use to boot from, so we simply put MBR + /boot on every disk. We had "make bootable" button, but users do mistakes and choose wrong disks, ending up with non-booting nodes.

> Basically this mean that disk management system in UI not working at all.
please be concrete, this statement has 0 info of what is broken

If user wants exclude some disks from Fuel configuration, and he is sure that those excluded are not the first to boot in BIOS settings, then it is another story - for which we rather need a blueprint and further discussion.

[Evgeniy L (rustyrobot)]

So, does this ticket have High status or not? We can use checkboxes for each disk (we had a discussion in mailing list about that https://<email address hidden>/msg00365.html).

This task will require a lot of time for testing (because disk allocation logic is complicated) as result we need to make decision about priority of this task.

[Mike Scherbakov (mihgen)]

I'm setting status to Invalid. Please feel free to reopen and raise this question in fuel-dev ML, however it is in full accordance with design document we had for disk partitioning back in 3.2.

[Aleksandr Shaposhnikov (alashai8)]

Customer would like to have disks in system that not used by openstack/ceph be untouched.

Real life example (consider this as extend for bug):
In fuel node's disk management section I have partitioned only sda drive. I also have sdb and sdc and leaved them unused because I plan to use them for my own on these servers. Maybe I have a backups/recovery on them or plan to use this SSD drives as b-cache. Whatever. Just leaved them unused in terms of fuel and start the deployment.

Expected behavior:
Openstack cluster is installed without any impact on unused disks. They should be in state that they've been before openstack provision/deployment.

Observed behavior:
Fuel is formatting them and create boot partition on each of them making raid 1 from all devices that I have.

[Evgeniy L (rustyrobot)]

Here is another workaround

I had 3 disks sda, sdb and sdc then I removed boot section via CLI

    fuel provisioning --env-id 1 --default -d

    # remove this section for sdc
    - size: 300
      type: boot
    - file_system: ext2
      mount: /boot
      name: Boot
      size: 200
      type: raid

    fuel provisioning --env-id 1 -u

    # run deployment

And there was no raid partition on sdc

    [root@node-2 ~]# cat /proc/mdstat
    Personalities : [raid1]
    md0 : active raid1 sda3[0] sdb3[1]
          204736 blocks super 1.0 [2/2] [UU]

For the next release we decided to make a special checkbox to not allocate raid on all disks.
Here is blueprint https://blueprints.launchpad.net/fuel/+spec/nailgun-bootable-checkbox-for-disks

[Meg McRoberts (dreidellhasa)]

Added to Known Issues in 5.0 Release Notes.

[Meg McRoberts (dreidellhasa)]

Listed as "Known Issue" in 5.0.1 Release Notes.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/133112