TestEncryptedCinderVolumes cryptsetup name is too long

[Mike Perez]

First off, while I understand this is not reproducible with the reference implementation LVM, this seems like a unknown limitation today since we're not enforcing any length on the IQN or recommending anything.

When running Datera storage with Cinder and the following TestEncryptedCinderVolumes tests:

{0} tempest.scenario.test_encrypted_cinder_volumes.TestEncryptedCinderVolumes.test_encrypted_cinder_volumes_cryptsetup
{0} tempest.scenario.test_encrypted_cinder_volumes.TestEncryptedCinderVolumes.test_encrypted_cinder_volumes_luks

cryptsetup complains about the name being too long:

http://paste.openstack.org/show/192537

Nova uses the device name that's in /dev/disk-by-path, which in this case is the returned iqn from the backend:

ip-172.30.128.2:3260-iscsi-iqn.2013-05.com.daterainc:OpenStack-TestEncryptedCinderVolumes-676292884:01:sn:aef6a6f1cd84768f-lun-0

Already started talking Matt Treinish about this on IRC last week. Unsure where the fix should actual go into.

[John Griffith (john-griffith)]

This seems to be causing a lot of confusion for people; The issue is only being seen by Datera right now because of the way it forms it's IQN using the display-name of the volume which is ridiculously long in this test case.

This should probably use the UUID of the volume and a prefix like we do in the LVM driver and others. Especially since we don't enforce unique display-names.

[Mike Perez (thingee)]

Going to take John's suggestion of just passing a uuid instead of the volume name in the iqn.

[Ramy Asselin (ramy-asselin)]

I think this should be re-opened because the iqn is within the specified limits.

"Initiators and targets MUST support the receipt of iSCSI names of up
   to the maximum length of 223 bytes." [1]

[1] https://www.ietf.org/rfc/rfc3720.txt 3.2.6.1. iSCSI Name Properties

[Walt Boring (walter-boring)]

For what it's worth, HP Lefthand Volume IQN's are just as long as Datera's. Cinder/Openstack cannot enforce length restriction on iSCSI array vendor's IQN length.

[Rajini Karthik (rajini-karthik)]

This bug affects Dell EQL driver as well. Can we reopen this bug?

[Sean McGinnis (sean-mcginnis)]

Dell - Internal Use - Confidential

We had to skip these tests in our CI as well. Our EqualLogic array's IQNs are too long as well.

-----Original Message-----
From: <email address hidden> [mailto:<email address hidden>] On Behalf Of Walt Boring
Sent: Tuesday, April 07, 2015 2:12 PM
To: McGinnis, Sean
Subject: [Bug 1432490] Re: TestEncryptedCinderVolumes cryptsetup name is too long

For what it's worth, HP Lefthand Volume IQN's are just as long as
Datera's. Cinder/Openstack cannot enforce length restriction on iSCSI
array vendor's IQN length.

--
You received this bug notification because you are subscribed to a
duplicate bug report (1440187).
https://bugs.launchpad.net/bugs/1432490

Title:
  TestEncryptedCinderVolumes cryptsetup name is too long

Status in Cinder:
  Invalid
Status in OpenStack Compute (Nova):
  Invalid
Status in Tempest:
  Invalid

Bug description:
  First off, while I understand this is not reproducible with the
  reference implementation LVM, this seems like a unknown limitation
  today since we're not enforcing any length on the IQN or recommending
  anything.

  When running Datera storage with Cinder and the following
  TestEncryptedCinderVolumes tests:

  {0} tempest.scenario.test_encrypted_cinder_volumes.TestEncryptedCinderVolumes.test_encrypted_cinder_volumes_cryptsetup
  {0} tempest.scenario.test_encrypted_cinder_volumes.TestEncryptedCinderVolumes.test_encrypted_cinder_volumes_luks

  cryptsetup complains about the name being too long:

  http://paste.openstack.org/show/192537

  Nova uses the device name that's in /dev/disk-by-path, which in this
  case is the returned iqn from the backend:

  ip-172.30.128.2:3260-iscsi-iqn.2013-05.com.daterainc:OpenStack-
  TestEncryptedCinderVolumes-676292884:01:sn:aef6a6f1cd84768f-lun-0

  Already started talking Matt Treinish about this on IRC last week.
  Unsure where the fix should actual go into.

To manage notifications about this bug go to:
https://bugs.launchpad.net/cinder/+bug/1432490/+subscriptions

[Walt Boring (walter-boring)]

blkid -s UUID -o value /dev/sdb works, if the device has a UUID attribute set on it.

We might want to look at setting the block device's UUID prior to calling cryptsetup on it. Then we could use the UUID
instead of the iqn.