Uploaded image for project: 'OpenShift Container Platform (OCP) Strategy'
  1. OpenShift Container Platform (OCP) Strategy
  2. OCPSTRAT-711

LVM Storage: wipe local volumes before first use

    XMLWordPrintable

Details

    • False
    • Hide

      None

      Show
      None
    • False
    • 100
    • 100% 100%
    • Hide
      LVMS does not touch a disk their might be data on it (e.g. a file system etc). This is to avoid accidental loss of data. This behaviour can now be disabled using the field storage.deviceClasses.deviceSelector.forceWipeDevicesAndDestroyAllData
      within LVMCluster CRD. If enabled, LVMS will wipe the devices, specified under deviceSelector, before using them. USE WITH CAUTION!
      Show
      LVMS does not touch a disk their might be data on it (e.g. a file system etc). This is to avoid accidental loss of data. This behaviour can now be disabled using the field storage.deviceClasses.deviceSelector.forceWipeDevicesAndDestroyAllData within LVMCluster CRD. If enabled, LVMS will wipe the devices, specified under deviceSelector, before using them. USE WITH CAUTION!
    • Feature
    • Proposed
    • 0
    • 0
    • Program Call

    Description

      Feature Overview (aka. Goal Summary)  

      Currently, LVMS does not touch a disk if it suspects that their might still be data on it (e.g. partitions, filesystems, volume groups etc.). This is to protect customers from data loss. Customers would need to manually access the host and wipe the disks themself. This is inconvienent, esp. in automated test setups where a server might be in an unknown state and manual intervention is not wanted. 

      The goal of this feature is to provide an attribute in the LVMS CRD that admins can set to force wipe delete the selected disks.

      Goals (aka. expected user outcomes)

      be able to force wipe disks, but only when disks are defined explicitly via their paths.

      Requirements (aka. Acceptance Criteria):

      Change the LVMCluster CRD with an new, optional attribute 

      spec.storage.deviceClasses[n].deviceSelector.forceWipeDevicesAndDestroyAllData

      of type Boolean with default = false.

      If that attribute is set to true, each disk listed under the deviceSelector.paths|optionalPaths  attributes are wiped out using "wipefs --all --force <path>"

      The action is to be logged on a log level that is usually enabled in a standard deployment (e.g. warning or info).

      Out of Scope

      LVMS has an 'auto' mode where it grabs all suitable disks, i.e. when no deviceSelector is given, all unused disks are used. In this 'auto' mode, no wiping of disks is happening. It would be far too dangerous to accidentally wipe an important disk.

      Documentation Considerations

      Docs need to be updated, in the sections:

      • Creating a Logical Volume Manager cluster
      • LVM Storage reference YAML file

      Interoperability Considerations

      Local Storage Operator is going to implement a similiar attribute. We should align / standardize on the exact name of that attribute. It should really point out explicititly that this might induce loss of data.

      Size

      Eng: M - API change, plumbing the functionality through the API to the CSI driver.

      Docs: M - Gigantic warning in the docs. Existing examples need modification to include the API change.

      QE: L - Requires many testing permutations. Will need to test how this works when software RAID is configured. Full regression testing on SNO and multi-node footprints.

      Attachments

        Issue Links

          Activity

            People

              dfroehli42rh Daniel Fröhlich
              dfroehli42rh Daniel Fröhlich
              Jakob Moeller, Suleyman Akbas
              Suleyman Akbas Suleyman Akbas
              Mike Fiedler Mike Fiedler
              Daniel Macpherson Daniel Macpherson
              Chad Scribner Chad Scribner
              Daniel Fröhlich Daniel Fröhlich
              Jon Thomas Jon Thomas
              Votes:
              0 Vote for this issue
              Watchers:
              5 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: