Add attaching state for Volumes

bug 897726 Add "attaching" status for volumes and clear state on error
This closes a gap where multiple calls of attach for the same volume
can happen before the volume is actually attached.

Change-Id: I59d22ceda83729c1a455af9994c9ffec1912e23b

7 files changed, 38 insertions, 5 deletions

diff --git a/Authors b/Authors
index ad6351b07..888f3008f 100644
--- a/Authors
+++ b/Authors
@@ -106,6 +106,7 @@ Kevin L. Mitchell <kevin.mitchell@rackspace.com>
 Kiall Mac Innes <kiall@managedit.ie>
 Kirill Shileev <kshileev@gmail.com>
 Koji Iida <iida.koji@lab.ntt.co.jp>
+Liam Kelleher  <liam.kelleher@hp.com>
 Likitha Shetty <likitha.shetty@citrix.com>
 Loganathan Parthipan <parthipan@hp.com>
 Lorin Hochstein <lorin@isi.edu>
diff --git a/nova/compute/api.py b/nova/compute/api.py
index edadf1978..874dcae47 100644
--- a/nova/compute/api.py
+++ b/nova/compute/api.py
@@ -1594,6 +1594,7 @@ class API(base.Base):
             raise exception.InvalidDevicePath(path=device)
         volume = self.volume_api.get(context, volume_id)
         self.volume_api.check_attach(context, volume)
+        self.volume_api.reserve_volume(context, volume)
         params = {"volume_id": volume_id,
                   "mountpoint": device}
         self._cast_compute_message('attach_volume', context,
diff --git a/nova/compute/manager.py b/nova/compute/manager.py
index f31a0d2e4..ee392f3e8 100644
--- a/nova/compute/manager.py
+++ b/nova/compute/manager.py
@@ -1647,12 +1647,21 @@ class ComputeManager(manager.SchedulerDependentManager):
         context = context.elevated()
         instance_ref = self.db.instance_get_by_uuid(context, instance_uuid)
         instance_id = instance_ref['id']
+        msg = _("instance %(instance_uuid)s: attaching volume %(volume_id)s"
+                " to %(mountpoint)s")
         LOG.audit(_('Attaching volume %(volume_id)s to %(mountpoint)s'),
                   locals(), context=context, instance=instance_ref)
-        connector = self.driver.get_volume_connector(instance_ref)
-        connection_info = self.volume_api.initialize_connection(context,
-                                                                volume,
-                                                                connector)
+        try:
+            connector = self.driver.get_volume_connector(instance_ref)
+            connection_info = self.volume_api.initialize_connection(context,
+                                                                    volume,
+                                                                    connector)
+        except Exception:  # pylint: disable=W0702
+            with utils.save_and_reraise_exception():
+                msg = _("instance %(instance_uuid)s: attach failed"
+                        " %(mountpoint)s, removing")
+                LOG.exception(msg % locals(), context=context)
+                self.volume_api.unreserve_volume(context, volume)
         try:
             self.driver.attach_volume(connection_info,
                                       instance_ref['name'],
diff --git a/nova/tests/policy.json b/nova/tests/policy.json
index c231bff60..d2e647f9f 100644
--- a/nova/tests/policy.json
+++ b/nova/tests/policy.json
@@ -127,6 +127,8 @@
 
     "volume:attach": [],
     "volume:detach": [],
+    "volume:reserve_volume": [],
+    "volume:unreserve_volume": [],
     "volume:check_attach": [],
     "volume:check_detach": [],
     "volume:initialize_connection": [],
diff --git a/nova/tests/test_compute.py b/nova/tests/test_compute.py
index 30027eb67..f8f87cba7 100644
--- a/nova/tests/test_compute.py
+++ b/nova/tests/test_compute.py
@@ -3049,12 +3049,16 @@ class ComputeAPITestCase(BaseTestCase):
         def fake_check_attach(*args, **kwargs):
             pass
 
+        def fake_reserve_volume(*args, **kwargs):
+            pass
+
         def fake_volume_get(self, context, volume_id):
             return {'id': volume_id}
 
         self.stubs.Set(nova.volume.api.API, 'get', fake_volume_get)
         self.stubs.Set(nova.volume.api.API, 'check_attach', fake_check_attach)
-
+        self.stubs.Set(nova.volume.api.API, 'reserve_volume',
+                       fake_reserve_volume)
         instance = self._create_fake_instance()
         self.compute_api.attach_volume(self.context, instance, 1, '/dev/vdb')
 
diff --git a/nova/utils.py b/nova/utils.py
index 85f39dbe0..172b68afa 100644
--- a/nova/utils.py
+++ b/nova/utils.py
@@ -24,6 +24,7 @@ import datetime
 import functools
 import hashlib
 import inspect
+import itertools
 import json
 import lockfile
 import os
@@ -712,6 +713,11 @@ def to_primitive(value, convert_instances=False, level=0):
         if test(value):
             return unicode(value)
 
+    # value of itertools.count doesn't get caught by inspects
+    # above and results in infinite loop when list(value) is called.
+    if type(value) == itertools.count:
+        return unicode(value)
+
     # FIXME(vish): Workaround for LP bug 852095. Without this workaround,
     #              tests that raise an exception in a mocked method that
     #              has a @wrap_exception with a notifier will fail. If
diff --git a/nova/volume/api.py b/nova/volume/api.py
index 8646f65e9..4719f4f1e 100644
--- a/nova/volume/api.py
+++ b/nova/volume/api.py
@@ -229,6 +229,15 @@ class API(base.Base):
                            'volume_id': volume['id']}})
 
     @wrap_check_policy
+    def reserve_volume(self, context, volume):
+        self.update(context, volume, {"status": "attaching"})
+
+    @wrap_check_policy
+    def unreserve_volume(self, context, volume):
+        if volume['status'] == "attaching":
+            self.update(context, volume, {"status": "available"})
+
+    @wrap_check_policy
     def attach(self, context, volume, instance_id, mountpoint):
         host = volume['host']
         queue = self.db.queue_get_for(context, FLAGS.volume_topic, host)
@@ -257,6 +266,7 @@ class API(base.Base):
 
     @wrap_check_policy
     def terminate_connection(self, context, volume, connector):
+        self.unreserve_volume(context, volume)
         host = volume['host']
         queue = self.db.queue_get_for(context, FLAGS.volume_topic, host)
         return rpc.call(context, queue,