qa: ignore trimmed cache items for dead cache drop

Fixes: https://tracker.ceph.com/issues/42986
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

diff --git a/qa/suites/fs/multiclient/tasks/cephfs_misc_tests.yaml b/qa/suites/fs/multiclient/tasks/cephfs_misc_tests.yaml
index 90ae385897a98b2129dcc2242f64bacbfa6a52d0..564989d6d766429d4f2e5a3b64c4f1c84e9aa0e3 100644 (file)
--- a/qa/suites/fs/multiclient/tasks/cephfs_misc_tests.yaml
+++ b/qa/suites/fs/multiclient/tasks/cephfs_misc_tests.yaml
@@ -9,3 +9,5 @@ overrides:
       - evicting unresponsive client
       - POOL_APP_NOT_ENABLED
       - has not responded to cap revoke by MDS for over
+      - MDS_CLIENT_LATE_RELEASE
+      - responding to mclientcaps

diff --git a/qa/tasks/cephfs/test_misc.py b/qa/tasks/cephfs/test_misc.py
index b21faa423b01964841e4d1f4d4caf9f64d4bd633..342fb76d4dbf535ae92d5e19edf1b669d32c2016 100644 (file)
--- a/qa/tasks/cephfs/test_misc.py
+++ b/qa/tasks/cephfs/test_misc.py
@@ -257,7 +257,12 @@ class TestCacheDrop(CephFSTestCase):
         self.assertEqual(result['flush_journal']['return_code'], 0)
         self.assertGreater(result['duration'], 5)
         self.assertLess(result['duration'], 120)
-        self.assertEqual(0, result['trim_cache']['trimmed'])
+        # Note: result['trim_cache']['trimmed'] may be >0 because dropping the
+        # cache now causes the Locker to drive eviction of stale clients (a
+        # stale session will be autoclosed at mdsmap['session_timeout']). The
+        # particular operation causing this is journal flush which causes the
+        # MDS to wait wait for cap revoke.
+        #self.assertEqual(0, result['trim_cache']['trimmed'])
         self.mount_a.kill_cleanup()
         self.mount_a.mount()
         self.mount_a.wait_until_mounted()
@@ -275,7 +280,11 @@ class TestCacheDrop(CephFSTestCase):
         self.assertEqual(result['flush_journal']['return_code'], 0)
         self.assertGreater(result['duration'], 5)
         self.assertLess(result['duration'], 120)
-        self.assertEqual(0, result['trim_cache']['trimmed'])
+        # Note: result['trim_cache']['trimmed'] may be >0 because dropping the
+        # cache now causes the Locker to drive eviction of stale clients (a
+        # stale session will be autoclosed at mdsmap['session_timeout']). The
+        # particular operation causing this is journal flush which causes the
+        # MDS to wait wait for cap revoke.
         self.mount_a.kill_cleanup()
         self.mount_a.mount()
         self.mount_a.wait_until_mounted()