RBD: add support for revert-to-snapshot

We note here that although this operation is implemented in the
Ceph backend, the Ceph docs indicate that it is inefficient and the
preferred method of returning to a known previous state is to clone
from a snapshot.  However, this is properly a backend operation, and
it does not make sense to try to do it better in the driver than it
can be done on the backend.

Co-authored-by: Brian Rosmaita <rosmaita.fossdev@gmail.com>
Implements: bp rbd-revert-to-snapshot
Change-Id: If8a5eb3a03e18f9043ff29f7648234c9b46376a0

.zuul.yaml

@@ -132,6 +132,11 @@
     vars:
       zuul_additional_subunit_dirs:
         - "{{ ansible_user_dir }}/{{ zuul.projects['opendev.org/openstack/cinderlib'].src_dir }}"
+      devstack_local_conf:
+        test-config:
+          $TEMPEST_CONFIG:
+            volume-feature-enabled:
+              volume_revert: True
 
 - job:
     name: cinder-grenade-dsvm-mn-sub-bak

cinder/tests/unit/volume/drivers/test_rbd.py

@@ -960,6 +960,17 @@ class RBDTestCase(test.TestCase):
                 self.assertTrue(proxy.unprotect_snap.called)
                 self.assertFalse(proxy.remove_snap.called)
 
+    @common_mocks
+    def test_snapshot_revert_use_temp_snapshot(self):
+        self.assertFalse(self.driver.snapshot_revert_use_temp_snapshot())
+
+    @common_mocks
+    def test_revert_to_snapshot(self):
+        image = self.mock_proxy.return_value.__enter__.return_value
+        self.driver.revert_to_snapshot(self.context, self.volume_a,
+                                       self.snapshot)
+        image.rollback_to_snap.assert_called_once_with(self.snapshot.name)
+
     @common_mocks
     def test_get_children_info(self):
         volume = self.mock_proxy

cinder/volume/drivers/rbd.py

@@ -1232,6 +1232,32 @@ class RBDDriver(driver.CloneableImageVD, driver.MigrateVD,
                 LOG.info("Snapshot %s does not exist in backend.",
                          snap_name)
 
+    def snapshot_revert_use_temp_snapshot(self):
+        """Disable the use of a temporary snapshot on revert."""
+        return False
+
+    def revert_to_snapshot(self, context, volume, snapshot):
+        """Revert a volume to a given snapshot."""
+        # NOTE(rosmaita): The Ceph documentation notes that this operation is
+        # inefficient on the backend for large volumes, and that the preferred
+        # method of returning to a pre-existing state in Ceph is to clone from
+        # a snapshot.
+        # So why don't we do something like that here?
+        # (a) an end user can do the more efficient operation on their own if
+        #     they value speed over the convenience of reverting their existing
+        #     volume
+        # (b) revert-to-snapshot is properly a backend operation, and should
+        #     be handled by the backend -- trying to "fake it" in this driver
+        #     is both dishonest and likely to cause subtle bugs
+        # (c) the Ceph project undergoes continual improvement.  It may be
+        #     the case that there are things an operator can do on the Ceph
+        #     side (for example, use BlueStore for the Ceph backend storage)
+        #     to improve the efficiency of this operation.
+        # Thus, a motivated operator reading this is encouraged to consult
+        # the Ceph documentation.
+        with RBDVolumeProxy(self, volume.name) as image:
+            image.rollback_to_snap(snapshot.name)
+
     def _disable_replication(self, volume):
         """Disable replication on the given volume."""
         vol_name = utils.convert_str(volume.name)

doc/source/reference/support-matrix.ini

@@ -775,7 +775,7 @@ driver.nimble=missing
 driver.pure=missing
 driver.qnap=missing
 driver.quobyte=missing
-driver.rbd=missing
+driver.rbd=complete
 driver.seagate=missing
 driver.storpool=missing
 driver.synology=missing

releasenotes/notes/rbd-support-revert-to-snapshot-c9ca62c9efbabf5f.yaml

@@ -0,0 +1,36 @@
+---
+features:
+  - |
+    RBD driver: support added for reverting a volume to the most recent
+    snapshot taken.
+
+    Please be aware of the following known issues with this operation
+    and the Ceph storage backend:
+
+    * Rolling back a volume to a snapshot overwrites the current volume
+      with the data from the snapshot, and the time it takes to complete
+      this operation increases with the size of the volume.
+
+      It is faster to create a new volume from a snapshot.  You may
+      wish to recommend this option to your users whose use cases do not
+      strictly require revert-to-snapshot.
+
+    * The efficiency of revert-to-snapshot is also dependent upon the
+      Ceph storage backend in use, namely, whether or not BlueStore is
+      being used in your Ceph installation.
+
+    Please consult the Ceph documentation for details.
+issues:
+  - |
+    RBD driver: There are some known issues concerning the revert-to-snapshot
+    support added in this release.
+
+    * The time it takes to complete the revert-to-snapshot operation increases
+      with the size of the volume.  It is faster to create a new volume from
+      a snapshot.
+
+    * The efficiency of revert-to-snapshot depends upon the Ceph storage
+      backend in use, particularly whether or not BlueStore is being used
+      in your Ceph installation.
+
+    Please consult the Ceph documentation for details.