mgr/cephadm: ok-to-stop for mgr

depends on #38854

Signed-off-by: Daniel-Pivonka <dpivonka@redhat.com>

diff --git a/src/pybind/mgr/cephadm/services/cephadmservice.py b/src/pybind/mgr/cephadm/services/cephadmservice.py
index d9fee1270ab0bf534f44f6fc420416f78a3f004c..da298b3e156379f2ca3d3bbbb6e71027281872d5 100644 (file)
--- a/src/pybind/mgr/cephadm/services/cephadmservice.py
+++ b/src/pybind/mgr/cephadm/services/cephadmservice.py
@@ -1,3 +1,4 @@
+import errno
 import json
 import re
 import logging
@@ -231,7 +232,7 @@ class CephadmService(metaclass=ABCMeta):
         logger.info(out)
         return HandleCommandResult(r.retval, out, r.stderr)
 
-    def _enough_daemons_to_stop(self, daemon_type: str, daemon_ids: List[str], service: str, low_limit: int) -> Tuple[bool, str]:
+    def _enough_daemons_to_stop(self, daemon_type: str, daemon_ids: List[str], service: str, low_limit: int, alert: bool = False) -> Tuple[bool, str]:
         # Provides a warning about if it possible or not to stop <n> daemons in a service
         names = [f'{daemon_type}.{d_id}' for d_id in daemon_ids]
         number_of_running_daemons = len(
@@ -247,9 +248,14 @@ class CephadmService(metaclass=ABCMeta):
         daemon_count = "only" if number_of_running_daemons == 1 else number_of_running_daemons
         left_count = "no" if num_daemons_left == 0 else num_daemons_left
 
-        out = (f'WARNING: Stopping {len(daemon_ids)} out of {number_of_running_daemons} daemons in {service} service. '
-               f'Service will not be operational with {left_count} {plural(num_daemons_left)} left. '
-               f'At least {low_limit} {plural(low_limit)} must be running to guarantee service. ')
+        if alert:
+            out = (f'ALERT: Cannot stop {names} in {service} service. '
+                   f'Not enough remaining {service} daemons. '
+                   f'Please deploy at least {low_limit + 1} {service} daemons before stopping {names}. ')
+        else:
+            out = (f'WARNING: Stopping {len(daemon_ids)} out of {number_of_running_daemons} daemons in {service} service. '
+                   f'Service will not be operational with {left_count} {plural(num_daemons_left)} left. '
+                   f'At least {low_limit} {plural(low_limit)} must be running to guarantee service. ')
         return True, out
 
     def pre_remove(self, daemon: DaemonDescription) -> None:
@@ -521,6 +527,21 @@ class MgrService(CephService):
         num = len(mgr_map.get('standbys'))
         return bool(num)
 
+    def ok_to_stop(self, daemon_ids: List[str], force: bool = False) -> HandleCommandResult:
+        # ok to stop if there is more than 1 mgr and not trying to stop the active mgr
+
+        warn, warn_message = self._enough_daemons_to_stop(self.TYPE, daemon_ids, 'Mgr', 1, True)
+        if warn:
+            return HandleCommandResult(-errno.EBUSY, '', warn_message)
+
+        mgr_daemons = self.mgr.cache.get_daemons_by_type(self.TYPE)
+        active = self.get_active_daemon(mgr_daemons).daemon_id
+        if active in daemon_ids:
+            warn_message = 'ALERT: Cannot stop active Mgr daemon, Please switch active Mgrs with \'ceph mgr fail %s\'' % active
+            return HandleCommandResult(-errno.EBUSY, '', warn_message)
+
+        return HandleCommandResult(0, warn_message, '')
+
 
 class MdsService(CephService):
     TYPE = 'mds'

diff --git a/src/pybind/mgr/cephadm/tests/test_upgrade.py b/src/pybind/mgr/cephadm/tests/test_upgrade.py
index 7d4f091e9c5fca70ddda147983ed6f84c34cefdf..420fbdd652c180c31c74f931667a77f9d5186efd 100644 (file)
--- a/src/pybind/mgr/cephadm/tests/test_upgrade.py
+++ b/src/pybind/mgr/cephadm/tests/test_upgrade.py
@@ -3,7 +3,7 @@ from unittest import mock
 
 import pytest
 
-from ceph.deployment.service_spec import ServiceSpec
+from ceph.deployment.service_spec import PlacementSpec, ServiceSpec
 from cephadm import CephadmOrchestrator
 from cephadm.upgrade import CephadmUpgrade
 from cephadm.serve import CephadmServe
@@ -33,78 +33,80 @@ def test_upgrade_start(cephadm_module: CephadmOrchestrator):
                              True
                          ])
 def test_upgrade_run(use_repo_digest, cephadm_module: CephadmOrchestrator):
-    with with_host(cephadm_module, 'test'):
-        cephadm_module.set_container_image('global', 'from_image')
-        if use_repo_digest:
-            cephadm_module.use_repo_digest = True
-        with with_service(cephadm_module, ServiceSpec('mgr'), CephadmOrchestrator.apply_mgr, 'test'),\
-             mock.patch("cephadm.module.CephadmOrchestrator.lookup_release_name",
-                        return_value='foo'),\
-             mock.patch("cephadm.module.CephadmOrchestrator.version",
-                        new_callable=mock.PropertyMock) as version_mock,\
-             mock.patch("cephadm.module.CephadmOrchestrator.get",
-                        return_value={
-                            # capture fields in both mon and osd maps
-                            "require_osd_release": "pacific",
-                            "min_mon_release": 16,
-                        }):
-            version_mock.return_value = 'ceph version 18.2.1 (somehash)'
-            assert wait(cephadm_module, cephadm_module.upgrade_start(
-                'to_image', None)) == 'Initiating upgrade to to_image'
-
-            assert wait(cephadm_module, cephadm_module.upgrade_status()).target_image == 'to_image'
-
-            def _versions_mock(cmd):
-                return json.dumps({
-                    'mgr': {
-                        'ceph version 1.2.3 (asdf) blah': 1
-                    }
-                })
-
-            cephadm_module._mon_command_mock_versions = _versions_mock
-
-            with mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(json.dumps({
-                'image_id': 'image_id',
-                'repo_digests': ['to_image@repo_digest'],
-                'ceph_version': 'ceph version 18.2.3 (hash)',
-            }))):
-
-                cephadm_module.upgrade._do_upgrade()
-
-            assert cephadm_module.upgrade_status is not None
-
-            with mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(
-                json.dumps([
-                    dict(
-                        name=list(cephadm_module.cache.daemons['test'].keys())[0],
-                        style='cephadm',
-                        fsid='fsid',
-                        container_id='container_id',
-                        container_image_id='image_id',
-                        container_image_digests=['to_image@repo_digest'],
-                        version='version',
-                        state='running',
-                    )
-                ])
-            )):
-                CephadmServe(cephadm_module)._refresh_hosts_and_daemons()
-
-            with mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(json.dumps({
-                'image_id': 'image_id',
-                'repo_digests': ['to_image@repo_digest'],
-                'ceph_version': 'ceph version 18.2.3 (hash)',
-            }))):
-                cephadm_module.upgrade._do_upgrade()
-
-            _, image, _ = cephadm_module.check_mon_command({
-                'prefix': 'config get',
-                'who': 'global',
-                'key': 'container_image',
-            })
+    with with_host(cephadm_module, 'host1'):
+        with with_host(cephadm_module, 'host2'):
+            cephadm_module.set_container_image('global', 'from_image')
             if use_repo_digest:
-                assert image == 'to_image@repo_digest'
-            else:
-                assert image == 'to_image'
+                cephadm_module.use_repo_digest = True
+            with with_service(cephadm_module, ServiceSpec('mgr', placement=PlacementSpec(host_pattern='*', count=2)), CephadmOrchestrator.apply_mgr, ''),\
+                mock.patch("cephadm.module.CephadmOrchestrator.lookup_release_name",
+                           return_value='foo'),\
+                mock.patch("cephadm.module.CephadmOrchestrator.version",
+                           new_callable=mock.PropertyMock) as version_mock,\
+                mock.patch("cephadm.module.CephadmOrchestrator.get",
+                           return_value={
+                               # capture fields in both mon and osd maps
+                               "require_osd_release": "pacific",
+                               "min_mon_release": 16,
+                           }):
+                version_mock.return_value = 'ceph version 18.2.1 (somehash)'
+                assert wait(cephadm_module, cephadm_module.upgrade_start(
+                    'to_image', None)) == 'Initiating upgrade to to_image'
+
+                assert wait(cephadm_module, cephadm_module.upgrade_status()
+                            ).target_image == 'to_image'
+
+                def _versions_mock(cmd):
+                    return json.dumps({
+                        'mgr': {
+                            'ceph version 1.2.3 (asdf) blah': 1
+                        }
+                    })
+
+                cephadm_module._mon_command_mock_versions = _versions_mock
+
+                with mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(json.dumps({
+                    'image_id': 'image_id',
+                    'repo_digests': ['to_image@repo_digest'],
+                    'ceph_version': 'ceph version 18.2.3 (hash)',
+                }))):
+
+                    cephadm_module.upgrade._do_upgrade()
+
+                assert cephadm_module.upgrade_status is not None
+
+                with mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(
+                    json.dumps([
+                        dict(
+                            name=list(cephadm_module.cache.daemons['host1'].keys())[0],
+                            style='cephadm',
+                            fsid='fsid',
+                            container_id='container_id',
+                            container_image_id='image_id',
+                            container_image_digests=['to_image@repo_digest'],
+                            version='version',
+                            state='running',
+                        )
+                    ])
+                )):
+                    CephadmServe(cephadm_module)._refresh_hosts_and_daemons()
+
+                with mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(json.dumps({
+                    'image_id': 'image_id',
+                    'repo_digests': ['to_image@repo_digest'],
+                    'ceph_version': 'ceph version 18.2.3 (hash)',
+                }))):
+                    cephadm_module.upgrade._do_upgrade()
+
+                _, image, _ = cephadm_module.check_mon_command({
+                    'prefix': 'config get',
+                    'who': 'global',
+                    'key': 'container_image',
+                })
+                if use_repo_digest:
+                    assert image == 'to_image@repo_digest'
+                else:
+                    assert image == 'to_image'
 
 
 def test_upgrade_state_null(cephadm_module: CephadmOrchestrator):