CKCT should only dispatch ubuntu_xfstests_* tests to nodes with scratch drives

[Po-Hsu Lin]

These ubuntu_xfstests_* test would require scratch drive to run, and it's super time consuming (3hr x 3 tests)

Therefore in KT we just run them on two specific nodes with amd64: bavor and rumford (the latter is not very reliable) and skip them on other arches

CKCT will dispatch jobs through the available node list. It will assign these jobs to a node that does not have a scratch drive. And the test will fail.

[Sean Feole (sfeole)]

There are 2 ways to do this, the correct way and the hack way.

Hack way == create a yaml and read from it to see which machines have /dev/sdb
Correct way == request the disk layouts of the machine under test and check to see if it has a second drive, if the machine does not, move onto the next machine.

[Po-Hsu Lin (cypressyew)]

For the correct way, we can use the following code in the control.ubuntu file from them, there is a paragraph of:
  # For right now, all of our (Ubuntu) testing makes use of a single,
  # scratch device.
  #
  if 'UBUNTU_SCRATCH_DEVICE' in argdict:
      device = argdict['UBUNTU_SCRATCH_DEVICE']
  else:
      device = '/dev/sdb'

Maybe we can use this to determine if a test needs a scratch drive.

[Po-Hsu Lin (cypressyew)]

Oh and one thing to note for the correct way, the node with a scratch drive will be assigned with more jobs (normal jobs + these huge xfstests) than other nodes, making it overloaded and time consuming to wait for normal jobs to run.

For the hacky way, in the lib/lab-configuration.yaml file we have this key:
    scratch drive: "/dev/sdb"

We can use this to create a new yaml if needed.