Merge lp:~codehelp/lava-scheduler/submitting-offline into lp:lava-scheduler

[Neil Williams]

Allow submission of jobs when devices are offline by only excluding devices which are retired from the calculation of available devices.

Currently built out on playground.v.l.o

[Neil Williams (codehelp)]

It works for singlenode jobs but something is still not right for multinode - I'm investigating.

review: Needs Fixing

[Neil Williams (codehelp)]

With the ~ negation, the device_types hash always returns 1, not the number of boards with a state other than the one specified.
On playground:
arndale 2 idle, 0 offline, 0 busy
kvm 0 idle, 2 offline, 0 busy
Submitting a MultiNode job for 2 KVMs gave:
Required number of device(s) 2 ([(u'arndale', 1), (u'kvm', 1)]) unavailable.
where the exception was changed to output:
"Required number of device(s) %d (%s) unavailable." % (count, device_types))

The alternative code (which adds OFFLINE to the original filter) is more messy and would need more ongoing maintenance when new states are added, but simple negation doesn't give us the results we expected.

    device_types = DeviceType.objects.values_list('name').filter(
        models.Q(device__status=Device.IDLE) |
        models.Q(device__status=Device.RUNNING) |
        models.Q(device__status=Device.RESERVED) |
        models.Q(device__status=Device.OFFLINE)
        ).annotate(
            num_count=models.Count('name')
        ).order_by('name')

With that change and deliberately testing with a MultiNode job which asks for 3 devices instead of 2:
Required number of device(s) 3 ([(u'arndale', 2), (u'kvm', 2)]) unavailable.

The state of the boards themselves did not change between these tests.

The original MultiNode job asking for 2 KVMs was submitted correctly on playground with the more expansive change.

I'll update the branch and this MP.

[Neil Williams (codehelp)]

Also changing the exception message to give something like:
Requested 3 kvm device(s) - only 2 available.

[lp:~codehelp/lava-scheduler/submitting-offline]

260. By Neil Williams on 2013-09-04

Change to an explicit check for boards which are offlining, reserved or
already offline to allow job submission as long as the boards
are not retired. Clarify exception message.

[Neil Williams (codehelp)]

review: Needs Resubmitting

[Antonio Terceiro (terceiro)]

Looks good to me

review: Approve

[Neil Williams (codehelp)]

review: Approve

=== modified file 'lava_scheduler_app/models.py'
--- lava_scheduler_app/models.py	2013-09-02 15:42:27 +0000
+++ lava_scheduler_app/models.py	2013-09-04 14:48:56 +0000
@@ -62,7 +62,9 @@
     device_types = DeviceType.objects.values_list('name').filter(
         models.Q(device__status=Device.IDLE) |
         models.Q(device__status=Device.RUNNING) |
-        models.Q(device__status=Device.RESERVED)
+        models.Q(device__status=Device.RESERVED) |
+        models.Q(device__status=Device.OFFLINE) |
+        models.Q(device__status=Device.OFFLINING)
         ).annotate(
             num_count=models.Count('name')
         ).order_by('name')
@@ -79,7 +81,7 @@
                 continue
             else:
                 raise DevicesUnavailableException(
-                    "Required number of device(s) unavailable.")
+                    "Requested %d %s device(s) - only %d available." % (count, board, all_devices[board]))
     return True