[RFE] During cleaning, remove also disk labels (not just partitions)

[John Fulton]

In addition to wipefs [0], could Ironic disk_utils also call `sgdisk -Z` or similar?

This issue affects TripleO Ceph deployments as `ceph-disk prepare` is called which in turn calls `sgdisk --mbrtogpt`. While testing a 1020 disk deployment with TripleO, when `sgdisk --mbrtogpt` was called it failed for 143 of the disks but succeeded for the other 877 of the disks; all 1020 of the disks had gone through Ironic cleaning [1]. The 143 disks all had unrecognised disk labels and the issue was worked around by labeling the disks and resuming the deployment to get all 1020 disks activated. I've verified also that `ceph-disk prepare` works with disks which have had `sgdisk -Z` run on them. Using something like `ceph-disk zap` in a preboot script can achieve the same effect, but it seems preferable for TripleO users to just let the Ironic cleaning take care of it.

AFAICT, `wipefs --all` removes partitions but not disk labels. Is there any reason why Ironic cleaning should leave pre-existing GPT/MBR (or even EFI) labels on a disk if the user already opted to have the disk cleaned?

[0] https://github.com/openstack/ironic-lib/commit/042aa9ab5a27e251c8fb2f1855695cf5e791ecf5
[1] As implemented in https://bugs.launchpad.net/ironic/+bug/1603411

[Vladyslav Drok (vdrok)]

I'm +1 on this.

[Pavlo Shchelokovskyy (pshchelo)]

I stumbled on something similar when implementing partitioning support in ansible-deploy driver, which is 'erase_devices_metadata' clean step is not thorough enough, destroying only partition tables, while some partition info is left in the headers of partitions themselves, and can interfere with future partitions if they happen to fall on the same start position.

As example of the logic I used to work this around in ansible-deploy please see this gerrit change

https://review.openstack.org/#/c/424291

I think it would be quite straightforward to extend logic of erase_devices_metadata in IPA in a similar way.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/469155

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/469155
Committed: https://git.openstack.org/cgit/openstack/ironic-lib/commit/?id=4491a52c904432d6ec48fd2751a2b1931223cb28
Submitter: Jenkins
Branch: master

commit 4491a52c904432d6ec48fd2751a2b1931223cb28
Author: Dmitry Tantsur <email address hidden>
Date: Tue May 30 17:05:22 2017 +0200

    Add 'sgdisk -Z' to destroy_disk_metadata

    Apparently, wipefs still leaves some metadata around, which can break ceph.
    Use sgdisk's zapping feature to be absolutely sure.

    Change-Id: I51aea8fb5ba02d63a05a3794405caf000ff28e75
    Closes-Bug: #1690458

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/ironic-lib 2.8.0 release.