mgr/cephadm: skip ok-to-stop for mons in upgrade if < 3 mons

Fixes: https://tracker.ceph.com/issues/50443
Signed-off-by: Adam King <adking@redhat.com>
(cherry picked from commit 4c7bfdbdd133820a1c0e8d129980ed810a77168d)

diff --git a/src/pybind/mgr/cephadm/tests/test_upgrade.py b/src/pybind/mgr/cephadm/tests/test_upgrade.py
index 0751192df0d04fb65287663bc4c548e90123a0f3..c4d6c27a01e926d8e227393b896fb6ba7b5c08ef 100644 (file)
--- a/src/pybind/mgr/cephadm/tests/test_upgrade.py
+++ b/src/pybind/mgr/cephadm/tests/test_upgrade.py
@@ -128,3 +128,17 @@ def test_not_enough_mgrs(cephadm_module: CephadmOrchestrator):
         with with_service(cephadm_module, ServiceSpec('mgr', placement=PlacementSpec(count=1)), CephadmOrchestrator.apply_mgr, ''):
             with pytest.raises(OrchestratorError):
                 wait(cephadm_module, cephadm_module.upgrade_start('image_id', None))
+
+
+@mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm('{}'))
+@mock.patch("cephadm.CephadmOrchestrator.check_mon_command")
+def test_enough_mons_for_ok_to_stop(check_mon_command, cephadm_module: CephadmOrchestrator):
+    # only 2 monitors, not enough for ok-to-stop to ever pass
+    check_mon_command.return_value = (
+        0, '{"monmap": {"mons": [{"name": "mon.1"}, {"name": "mon.2"}]}}', '')
+    assert not cephadm_module.upgrade._enough_mons_for_ok_to_stop()
+
+    # 3 monitors, ok-to-stop should work fine
+    check_mon_command.return_value = (
+        0, '{"monmap": {"mons": [{"name": "mon.1"}, {"name": "mon.2"}, {"name": "mon.3"}]}}', '')
+    assert cephadm_module.upgrade._enough_mons_for_ok_to_stop()

diff --git a/src/pybind/mgr/cephadm/upgrade.py b/src/pybind/mgr/cephadm/upgrade.py
index 6ac76b4c8523b818cd682fd5c9490c8edf527189..830b1cb1a57b9de8df2bfc75b886443118d4c18f 100644 (file)
--- a/src/pybind/mgr/cephadm/upgrade.py
+++ b/src/pybind/mgr/cephadm/upgrade.py
@@ -420,6 +420,19 @@ class CephadmUpgrade:
 
         return continue_upgrade
 
+    def _enough_mons_for_ok_to_stop(self) -> bool:
+        # type () -> bool
+        ret, out, err = self.mgr.check_mon_command({
+            'prefix': 'quorum_status',
+        })
+        try:
+            j = json.loads(out)
+        except Exception:
+            raise OrchestratorError('failed to parse quorum status')
+
+        mons = [m['name'] for m in j['monmap']['mons']]
+        return len(mons) > 2
+
     def _do_upgrade(self):
         # type: () -> None
         if not self.upgrade_state:
@@ -567,12 +580,16 @@ class CephadmUpgrade:
                         to_upgrade.append(d_entry)
                     continue
 
-                if d.daemon_type in ['mon', 'osd', 'mds']:
+                if d.daemon_type in ['osd', 'mds']:
                     # NOTE: known_ok_to_stop is an output argument for
                     # _wait_for_ok_to_stop
                     if not self._wait_for_ok_to_stop(d, known_ok_to_stop):
                         return
 
+                if d.daemon_type == 'mon' and self._enough_mons_for_ok_to_stop():
+                    if not self._wait_for_ok_to_stop(d, known_ok_to_stop):
+                        return
+
                 to_upgrade.append(d_entry)
 
                 # if we don't have a list of others to consider, stop now