mgr/cephadm: block draining last _admin host

Fixes: https://tracker.ceph.com/issues/54413
Signed-off-by: Adam King <adking@redhat.com>
(cherry picked from commit a0d21c7108e8f95b541bdb7653d2595f68e42520)

diff --git a/src/pybind/mgr/cephadm/module.py b/src/pybind/mgr/cephadm/module.py
index 7aac8fb23e48943dbadf08bc627cbe505c3bacca..133f574c0b27ce97eb8f3a035f6e8a2bb72eb35c 100644 (file)
--- a/src/pybind/mgr/cephadm/module.py
+++ b/src/pybind/mgr/cephadm/module.py
@@ -2713,12 +2713,26 @@ Then run the following:
         return self.to_remove_osds.all_osds()
 
     @handle_orch_error
-    def drain_host(self, hostname):
-        # type: (str) -> str
+    def drain_host(self, hostname, force=False):
+        # type: (str, bool) -> str
         """
         Drain all daemons from a host.
         :param host: host name
         """
+
+        # if we drain the last admin host we could end up removing the only instance
+        # of the config and keyring and cause issues
+        if not force:
+            p = PlacementSpec(label='_admin')
+            admin_hosts = p.filter_matching_hostspecs(self.inventory.all_specs())
+            if len(admin_hosts) == 1 and admin_hosts[0] == hostname:
+                raise OrchestratorValidationError(f"Host {hostname} is the last host with the '_admin'"
+                                                  " label.\nDraining this host could cause the removal"
+                                                  " of the last cluster config/keyring managed by cephadm.\n"
+                                                  "It is recommended to add the _admin label to another host"
+                                                  " before completing this operation.\nIf you're certain this is"
+                                                  " what you want rerun this command with --force.")
+
         self.add_host_label(hostname, '_no_schedule')
 
         daemons: List[orchestrator.DaemonDescription] = self.cache.get_daemons_by_host(hostname)

diff --git a/src/pybind/mgr/orchestrator/_interface.py b/src/pybind/mgr/orchestrator/_interface.py
index 1f389448547b8f267ab23cf9c330fb07f2068bcc..1d7341c070165ce9b49b728b96de2d4b3ff287ae 100644 (file)
--- a/src/pybind/mgr/orchestrator/_interface.py
+++ b/src/pybind/mgr/orchestrator/_interface.py
@@ -355,7 +355,7 @@ class Orchestrator(object):
         """
         raise NotImplementedError()
 
-    def drain_host(self, hostname: str) -> OrchResult[str]:
+    def drain_host(self, hostname: str, force: bool = False) -> OrchResult[str]:
         """
         drain all daemons from a host
 

diff --git a/src/pybind/mgr/orchestrator/module.py b/src/pybind/mgr/orchestrator/module.py
index 88f24365294b56bb62bc863840b994aa0483c3c5..f33ecd50d021c34ac1041cbd9a396cc744d25224 100644 (file)
--- a/src/pybind/mgr/orchestrator/module.py
+++ b/src/pybind/mgr/orchestrator/module.py
@@ -360,9 +360,9 @@ class OrchestratorCli(OrchestratorClientMixin, MgrModule,
         return HandleCommandResult(stdout=completion.result_str())
 
     @_cli_write_command('orch host drain')
-    def _drain_host(self, hostname: str) -> HandleCommandResult:
+    def _drain_host(self, hostname: str, force: bool = False) -> HandleCommandResult:
         """drain all daemons from a host"""
-        completion = self.drain_host(hostname)
+        completion = self.drain_host(hostname, force)
         raise_if_exception(completion)
         return HandleCommandResult(stdout=completion.result_str())