qa: Add mirror test for snapshot with only dir

Fixes: https://tracker.ceph.com/issues/75804
Signed-off-by: Kotresh HR <khiremat@redhat.com>

diff --git a/qa/tasks/cephfs/test_mirroring.py b/qa/tasks/cephfs/test_mirroring.py
index 04fa198df4754958c0623e509d2c7d1f9573c4ae..dbe6dfb1fa7ddc3056d12cc3cdda56609b0fc382 100644 (file)
--- a/qa/tasks/cephfs/test_mirroring.py
+++ b/qa/tasks/cephfs/test_mirroring.py
@@ -1777,3 +1777,68 @@ class TestMirroring(CephFSTestCase):
         self.assertLess(d2_sync_time_stamp, d0_sync_time_stamp)
 
         self.disable_mirroring(self.primary_fs_name, self.primary_fs_id)
+
+    def test_cephfs_mirror_multithread_snapshot_only_dirs_sync(self):
+        """
+        Test that snapshot containing only directories sync and doesn't hang.
+
+        ... When snapshot containing only directories and no files, are queued for syncing, the sync hangs in following scenario.
+            1. Configure say /d0 and /d1 for mirroring.
+            2. Create around 10k files in /d0
+            3. Create a single dir say /d1/dir0 or nothing
+            4. snapshot /d0 and wait for status to change to 'syncing'
+            5. Now, snapshot /d1.
+
+        The /d1 snapshot will be stuck in syncing for ever.
+        See tracker https://tracker.ceph.com/issues/75804 for more details.
+        """
+        self.setup_mount_b(mds_perm='rw')
+        peer_spec = "client.mirror_remote@ceph"
+
+        # create 2 directories to snap
+        self.mount_a.run_shell(["mkdir", "d0"])
+        self.mount_a.run_shell(["mkdir", "d1"])
+
+        self.mount_a.create_n_files("d0/myfile", 10000)
+        self.mount_a.run_shell(["mkdir", "d1/dir0"])
+
+        log.debug('enabling mirroring')
+        self.enable_mirroring(self.primary_fs_name, self.primary_fs_id)
+        # The issue happens when the snapshot's syncm object is waiting for datasync threads
+        # at syncm_q and doesn't hit if syncm object is waiting at syncm's data queue for files.
+        # This can be achieved with the following.
+        #  1. Create more files (10k) for /d0 -
+        #       This provides enough time for /d1's crawl to finish before datasync threads picks
+        #       up /d1 for syncing. If the crawl is not finished when data sync threads are available,
+        #       it will go and wait for files at syncm's data queue.
+        #  2. Disable distribute_datasync_threads options. This guarantees that the /d1
+        #       is not picked up for syncing before it's crawling is finished and provides enough
+        #       window for the /d1's crawling to finish as all threads are busy syncing /d0's
+        #       10k files
+        log.debug('disabling cephfs_mirror_distribute_datasync_threads config')
+        self.config_set('client.mirror', 'cephfs_mirror_distribute_datasync_threads', 'false')
+
+        log.debug('adding directory paths')
+        self.add_directory(self.primary_fs_name, self.primary_fs_id, '/d0')
+        self.add_directory(self.primary_fs_name, self.primary_fs_id, '/d1')
+        self.peer_add(self.primary_fs_name, self.primary_fs_id, peer_spec, self.secondary_fs_name)
+
+        # take /d0 snapshot first, so that it starts syncing
+        log.debug('take /d0 snapshot first')
+        snap_name = "snap0"
+        self.mount_a.run_shell(["mkdir", f"d0/.snap/{snap_name}"])
+        log.debug('checking /d0/.snap/snap0 in progress')
+        self.check_peer_snap_in_progress(self.primary_fs_name, self.primary_fs_id,
+                                         peer_spec, '/d0', 'snap0')
+
+        # now that /d0 is in progress, take snaps of /d1
+        self.mount_a.run_shell(["mkdir", f"d1/.snap/{snap_name}"])
+
+        # Wait for d0, d1
+        self.check_peer_status(self.primary_fs_name, self.primary_fs_id, peer_spec, '/d0', 'snap0', 1)
+        self.verify_snapshot('d0', 'snap0')
+        self.check_peer_status(self.primary_fs_name, self.primary_fs_id, peer_spec, '/d1', 'snap0', 1)
+        self.verify_snapshot('d1', 'snap0')
+
+        self.config_set('client.mirror', 'cephfs_mirror_distribute_datasync_threads', 'true')
+        self.disable_mirroring(self.primary_fs_name, self.primary_fs_id)