Make scheduler check online_extend_support capability

Since Pike release, Cinder supports volume online extending, and by
default it assumes that every backend supports this feature. This
assumption causes a bug on those backends that don't support it. On
such backends, an online extending attempt leaves the volume in
error_extending state.

This patch allows a backend to report to the scheduler if it does not
support online extending. This way, an online extending attempt will
fail, without leaving the volume in error_extending state.

Closes-bug: #1765182
Change-Id: I2c31b5c171574074a8fc7ba86f94f983fc9658f7
Co-Authored-By: Lucio Seki <luciomitsuru.seki@fit-tecnologia.org.br>

cinder/interface/volume_driver.py

@@ -132,6 +132,9 @@ class VolumeDriverCore(base.CinderInterface):
         * sparse_copy_volume (Boolean)
             Whether copies performed by the volume manager for operations such
             as migration should attempt to preserve sparseness.
+        * online_extend_support (Boolean)
+            Whether the backend supports in-use volume extend or not. Defaults
+            to True.
 
         The returned dict may also contain a list, "pools", which has a similar
         dict for each pool being used with the backend.
@@ -255,6 +258,9 @@ class VolumeDriverCore(base.CinderInterface):
 
         :param volume: The volume to extend.
         :param new_size: The new desired size of the volume.
+
+        Note that if the volume backend doesn't support extending an in-use
+        volume, the driver should report online_extend_support=False.
         """
 
     def create_snapshot(self, snapshot):

cinder/scheduler/filters/capabilities_filter.py

@@ -15,6 +15,7 @@
 
 from oslo_log import log as logging
 
+from cinder.objects.fields import VolumeAttachStatus
 from cinder.scheduler import filters
 from cinder.scheduler.filters import extra_specs_ops
 
@@ -24,13 +25,27 @@ LOG = logging.getLogger(__name__)
 class CapabilitiesFilter(filters.BaseBackendFilter):
     """BackendFilter to work with resource (instance & volume) type records."""
 
-    def _satisfies_extra_specs(self, capabilities, resource_type):
+    def _satisfies_extra_specs(self, capabilities, filter_properties):
         """Check if capabilities satisfy resource type requirements.
 
         Check that the capabilities provided by the services satisfy
         the extra specs associated with the resource type.
         """
 
+        req_spec = filter_properties.get('request_spec')
+        if req_spec and req_spec.get('operation') == 'extend_volume':
+            # NOTE(erlon): By default, cinder considers that every backend
+            # supports volume online extending. Those backends that don't
+            # support it should report online_extend_support=False.
+            online_extends = capabilities.get('online_extend_support', True)
+            if online_extends is False:
+                vol_prop = req_spec.get('volume_properties')
+                attach_status = vol_prop.get('attach_status')
+                if attach_status != VolumeAttachStatus.DETACHED:
+                    LOG.debug("Backend doesn't support attached volume extend")
+                    return False
+
+        resource_type = filter_properties.get('resource_type')
         if not resource_type:
             return True
 
@@ -80,9 +95,8 @@ class CapabilitiesFilter(filters.BaseBackendFilter):
         # Note(zhiteng) Currently only Cinder and Nova are using
         # this filter, so the resource type is either instance or
         # volume.
-        resource_type = filter_properties.get('resource_type')
         if not self._satisfies_extra_specs(backend_state.capabilities,
-                                           resource_type):
+                                           filter_properties):
             LOG.debug("%(backend_state)s fails resource_type extra_specs "
                       "requirements", {'backend_state': backend_state})
             return False

cinder/tests/unit/scheduler/fakes.py

@@ -39,6 +39,7 @@ SERVICE_STATES = {
               'volume_backend_name': 'lvm1',
               'timestamp': UTC_NOW,
               'multiattach': True,
+              'online_extend_support': True,
               'uuid': 'a3a593da-7f8d-4bb7-8b4c-f2bc1e0b4824'},
     'host2': {'total_capacity_gb': 2048,
               'free_capacity_gb': 300,
@@ -50,6 +51,7 @@ SERVICE_STATES = {
               'reserved_percentage': 10,
               'volume_backend_name': 'lvm2',
               'timestamp': UTC_NOW,
+              'online_extend_support': False,
               'uuid': '4200b32b-0bf9-436c-86b2-0675f6ac218e'},
     'host3': {'total_capacity_gb': 512,
               'free_capacity_gb': 256,

cinder/tests/unit/scheduler/test_filter_scheduler.py

@@ -426,6 +426,42 @@ class FilterSchedulerTestCase(test_scheduler.SchedulerTestCase):
                           ctx, 'host1#lvm1', request_spec, {})
         self.assertTrue(_mock_service_get_topic.called)
 
+    @mock.patch('cinder.db.service_get_all')
+    def test_backend_passes_filters_online_extend_support_happy_day(
+            self, _mock_service_get_topic):
+        """Do a successful online extend with backend_passes_filters()."""
+        sched, ctx = self._backend_passes_filters_setup(
+            _mock_service_get_topic)
+        request_spec = {'volume_id': fake.VOLUME_ID,
+                        'volume_type': {'name': 'LVM_iSCSI'},
+                        'volume_properties': {'project_id': 1,
+                                              'size': 1,
+                                              'attach_status': 'attached'},
+                        'operation': 'extend_volume'}
+        request_spec = objects.RequestSpec.from_primitives(request_spec)
+        # host1#lvm1 has online_extend_support = True
+        sched.backend_passes_filters(ctx, 'host1#lvm1', request_spec, {})
+        self.assertTrue(_mock_service_get_topic.called)
+
+    @mock.patch('cinder.db.service_get_all')
+    def test_backend_passes_filters_no_online_extend_support(
+            self, _mock_service_get_topic):
+        """Fail the host due to lack of online extend support."""
+        sched, ctx = self._backend_passes_filters_setup(
+            _mock_service_get_topic)
+        request_spec = {'volume_id': fake.VOLUME_ID,
+                        'volume_type': {'name': 'LVM_iSCSI'},
+                        'volume_properties': {'project_id': 1,
+                                              'size': 1,
+                                              'attach_status': 'attached'},
+                        'operation': 'extend_volume'}
+        request_spec = objects.RequestSpec.from_primitives(request_spec)
+        # host2#lvm2 has online_extend_support = False
+        self.assertRaises(exception.NoValidBackend,
+                          sched.backend_passes_filters,
+                          ctx, 'host2#lvm2', request_spec, {})
+        self.assertTrue(_mock_service_get_topic.called)
+
     @mock.patch('cinder.db.service_get_all')
     def test_retype_policy_never_migrate_pass(self, _mock_service_get_topic):
         # Retype should pass if current host passes filters and

doc/source/contributor/drivers.rst

@@ -87,6 +87,11 @@ the backend cannot report the value or 'infinite' if the backend has no upper
 limit. But, it is recommended to report real values as the Cinder scheduler
 assigns lowest weight to any storage backend reporting 'unknown' or 'infinite'.
 
+**NOTE:** By default, Cinder assumes that the driver supports attached volume
+extending. If it doesn't, it should report 'online_extend_support=False'.
+Otherwise the scheduler will attempt to perform the operation, and may leave
+the volume in 'error_extending' state.
+
 Feature Enforcement
 -------------------
 

releasenotes/notes/bug-1765182-bcafd577f4b81eb6.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    Make Cinder scheduler check if backend reports `online_extend_support`
+    before performing an online extend operation.