Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-514

[OCPonRHV] CSI provisioned disks are effectively preallocated due to go-ovirt-client setting Provisioned and Initial size of the disk to the same value

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Done
    • Icon: Major Major
    • None
    • 4.11
    • None
    • Important
    • None
    • False
    • Hide

      None

      Show
      None

      When a thin provisioned COW format disk is created on OCP on RHV via CSI driver (a PVC -
      https://github.com/openshift/ovirt-csi-driver/blob/master/deploy/example/storage-claim.yaml

      • with for example requested storage 100 GB), the go-ovirt-client behaviour makes it so that the created disk has virtual size 100 GB and it's actual size is 110 GB.

      But this is thin provisioned disk, so the initial size of the disk should be default of the engine and then grow as needed, it shouldn't be this big.

      This causes all the disks created this way to be functionally preallocated (since it eats all that space), which is a real waste of space.

      How reproducible: 100%

      Steps to Reproduce:
      1. Create a storage claim (PVC) in Openshift (
      https://github.com/openshift/ovirt-csi-driver/blob/master/deploy/example/storage-claim.yaml
      ) using the default storage class (or any other storage class with thinProvisioning: "true") and with requested storage i.e. 100Gi

      $ oc create -f storage-claim.yaml

      2. In the RHV web console navigate to Storage -> Disks and check Virtual size and Actual size of the created disk (PVC)

      Actual results:
      Disk from our example with requested storage 100GB reports virtual size 100GB and actual size 110 GB.

      Expected results:
      Thin provisioned disks should start with small initial size and then grow as needed, so its actual size should be considerably smaller (the default initial size set by the engine should be 2.5 GB if I'm not mistaken).

      Note: The extra 10GB in the actual size are caused by overhead for the qcow2 disk format, which is 10%, and this was tracked here as a separate issue:
      https://bugzilla.redhat.com/show_bug.cgi?id=2097139

            vfuxova@redhat.com Veronika Fuxova (Inactive)
            vfuxova@redhat.com Veronika Fuxova (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated:
              Resolved: