mgr/volumes: Add clone failure reason in clone status

Add the clone failure reason in the clone status.
The sample output is as below:

$ ceph fs clone status cephfs clone_0
{
  "status": {
    "state": "failed",
    "source": {
      "volume": "cephfs",
      "subvolume": "subvolume_0",
      "snapshot": "snapshot_0",
      "size": "52428800"
    },
    "failure": {
      "errno": "2",
      "error_msg": "snapshot 'snapshot_0' does not exist"
    }
  }
}

Fixes: https://tracker.ceph.com/issues/55190
Signed-off-by: Kotresh HR <khiremat@redhat.com>

diff --git a/src/pybind/mgr/volumes/fs/async_cloner.py b/src/pybind/mgr/volumes/fs/async_cloner.py
index 810397031f52b22186b5e9bafe11b5350d5765c1..5a1fcbfdf5d006919211fe24e0d22e800b85a324 100644 (file)
--- a/src/pybind/mgr/volumes/fs/async_cloner.py
+++ b/src/pybind/mgr/volumes/fs/async_cloner.py
@@ -106,6 +106,8 @@ def handle_clone_pending(fs_client, volspec, volname, index, groupname, subvolna
             next_state = SubvolumeOpSm.transition(SubvolumeTypes.TYPE_CLONE,
                                                   SubvolumeStates.STATE_PENDING,
                                                   SubvolumeActions.ACTION_CANCELLED)
+            update_clone_failure_status(fs_client, volspec, volname, groupname, subvolname,
+                                        VolumeException(-errno.EINTR, "user interrupted clone operation"))
         else:
             next_state = SubvolumeOpSm.transition(SubvolumeTypes.TYPE_CLONE,
                                                   SubvolumeStates.STATE_PENDING,
@@ -184,7 +186,7 @@ def bulk_copy(fs_handle, source_path, dst_path, should_cancel):
                 raise VolumeException(-e.args[0], e.args[1])
     cptree(source_path, dst_path)
     if should_cancel():
-        raise VolumeException(-errno.EINTR, "clone operation interrupted")
+        raise VolumeException(-errno.EINTR, "user interrupted clone operation")
 
 def set_quota_on_clone(fs_handle, clone_volumes_pair):
     src_path = clone_volumes_pair[1].snapshot_data_path(clone_volumes_pair[2])
@@ -225,6 +227,14 @@ def do_clone(fs_client, volspec, volname, groupname, subvolname, should_cancel):
             bulk_copy(fs_handle, src_path, dst_path, should_cancel)
             set_quota_on_clone(fs_handle, clone_volumes)
 
+def update_clone_failure_status(fs_client, volspec, volname, groupname, subvolname, ve):
+    with open_volume_lockless(fs_client, volname) as fs_handle:
+        with open_clone_subvolume_pair(fs_client, fs_handle, volspec, volname, groupname, subvolname) as clone_volumes:
+            if ve.errno == -errno.EINTR:
+                clone_volumes[0].add_clone_failure(-ve.errno, "user interrupted clone operation")
+            else:
+                clone_volumes[0].add_clone_failure(-ve.errno, ve.error_str)
+
 def log_clone_failure(volname, groupname, subvolname, ve):
     if ve.errno == -errno.EINTR:
         log.info("Clone cancelled: ({0}, {1}, {2})".format(volname, groupname, subvolname))
@@ -240,6 +250,7 @@ def handle_clone_in_progress(fs_client, volspec, volname, index, groupname, subv
                                               SubvolumeStates.STATE_INPROGRESS,
                                               SubvolumeActions.ACTION_SUCCESS)
     except VolumeException as ve:
+        update_clone_failure_status(fs_client, volspec, volname, groupname, subvolname, ve)
         log_clone_failure(volname, groupname, subvolname, ve)
         next_state = get_next_state_on_error(ve.errno)
     except OpSmException as oe:
@@ -349,6 +360,7 @@ class Cloner(AsyncJobs):
                                                   clone_state,
                                                   SubvolumeActions.ACTION_CANCELLED)
             clone_subvolume.state = (next_state, True)
+            clone_subvolume.add_clone_failure(errno.EINTR, "user interrupted clone operation")
             s_subvolume.detach_snapshot(s_snapname, track_idx.decode('utf-8'))
 
     def cancel_job(self, volname, job):

diff --git a/src/pybind/mgr/volumes/fs/operations/versions/metadata_manager.py b/src/pybind/mgr/volumes/fs/operations/versions/metadata_manager.py
index 1b6c432783783f9b378a1b9409c51551880ec60c..942353926d6f4e28c6ef8f7782259e723186f7e8 100644 (file)
--- a/src/pybind/mgr/volumes/fs/operations/versions/metadata_manager.py
+++ b/src/pybind/mgr/volumes/fs/operations/versions/metadata_manager.py
@@ -26,6 +26,10 @@ class MetadataManager(object):
     GLOBAL_META_KEY_PATH    = "path"
     GLOBAL_META_KEY_STATE   = "state"
 
+    CLONE_FAILURE_SECTION = "CLONE_FAILURE"
+    CLONE_FAILURE_META_KEY_ERRNO = "errno"
+    CLONE_FAILURE_META_KEY_ERROR_MSG = "error_msg"
+
     MAX_IO_BYTES = 8 * 1024
 
     def __init__(self, fs, config_path, mode):

diff --git a/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v1.py b/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v1.py
index 42c08e04712b173fe2fbbd3a0db9c7d917f4728b..0e63b85878da9ad893233e3c854c903232a271ad 100644 (file)
--- a/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v1.py
+++ b/src/pybind/mgr/volumes/fs/operations/versions/subvolume_v1.py
@@ -139,6 +139,14 @@ class SubvolumeV1(SubvolumeBase, SubvolumeTemplate):
         if flush:
             self.metadata_mgr.flush()
 
+    def add_clone_failure(self, errno, error_msg):
+        self.metadata_mgr.add_section(MetadataManager.CLONE_FAILURE_SECTION)
+        self.metadata_mgr.update_section(MetadataManager.CLONE_FAILURE_SECTION,
+                                         MetadataManager.CLONE_FAILURE_META_KEY_ERRNO, errno)
+        self.metadata_mgr.update_section(MetadataManager.CLONE_FAILURE_SECTION,
+                                         MetadataManager.CLONE_FAILURE_META_KEY_ERROR_MSG, error_msg)
+        self.metadata_mgr.flush()
+
     def create_clone(self, pool, source_volname, source_subvolume, snapname):
         subvolume_type = SubvolumeTypes.TYPE_CLONE
         try:
@@ -660,6 +668,13 @@ class SubvolumeV1(SubvolumeBase, SubvolumeTemplate):
             raise VolumeException(-errno.EINVAL, "error fetching subvolume metadata")
         return clone_source
 
+    def _get_clone_failure(self):
+        clone_failure = {
+            'errno'     : self.metadata_mgr.get_option(MetadataManager.CLONE_FAILURE_SECTION, MetadataManager.CLONE_FAILURE_META_KEY_ERRNO),
+            'error_msg' : self.metadata_mgr.get_option(MetadataManager.CLONE_FAILURE_SECTION, MetadataManager.CLONE_FAILURE_META_KEY_ERROR_MSG),
+        }
+        return clone_failure
+
     @property
     def status(self):
         state = SubvolumeStates.from_value(self.metadata_mgr.get_global_option(MetadataManager.GLOBAL_META_KEY_STATE))
@@ -669,6 +684,12 @@ class SubvolumeV1(SubvolumeBase, SubvolumeTemplate):
         }
         if not SubvolumeOpSm.is_complete_state(state) and subvolume_type == SubvolumeTypes.TYPE_CLONE:
             subvolume_status["source"] = self._get_clone_source()
+        if SubvolumeOpSm.is_failed_state(state) and subvolume_type == SubvolumeTypes.TYPE_CLONE:
+            try:
+                subvolume_status["failure"] = self._get_clone_failure()
+            except MetadataMgrException:
+                pass
+
         return subvolume_status
 
     @property