mgr/cephadm: Daemons running of unreachable servers are offline

Fixes: https://tracker.ceph.com/issues/44602
Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>
(cherry picked from commit adca3e08e8e9f4a683f7c6cef8e96705cc340188)

diff --git a/src/pybind/mgr/cephadm/module.py b/src/pybind/mgr/cephadm/module.py
index d7b6b0f76509b5a2aee1d43c4187ca8e81e2a2fc..b730888e5c4eb51f82229eda03205d3f8913d6a0 100644 (file)
--- a/src/pybind/mgr/cephadm/module.py
+++ b/src/pybind/mgr/cephadm/module.py
@@ -2,7 +2,7 @@ import json
 import errno
 import logging
 import time
-import yaml
+from copy import copy
 from threading import Event
 from functools import wraps
 
@@ -320,6 +320,18 @@ class HostCache():
                 r.append(dd)
         return r
 
+    def get_daemons_with_volatile_status(self) -> Iterator[Tuple[str, Dict[str, orchestrator.DaemonDescription]]]:
+        for host, dm in self.daemons.items():
+            if host in self.mgr.offline_hosts:
+                def set_offline(dd: orchestrator.DaemonDescription) -> orchestrator.DaemonDescription:
+                    ret = copy(dd)
+                    ret.status = -1
+                    ret.status_desc = 'host is offline'
+                    return ret
+                yield host, {name: set_offline(d) for name, d in dm.items()}
+            else:
+                yield host, dm
+
     def get_daemons_by_service(self, service_name):
         # type: (str) -> List[orchestrator.DaemonDescription]
         result = []   # type: List[orchestrator.DaemonDescription]
@@ -1854,7 +1866,7 @@ class CephadmOrchestrator(orchestrator.Orchestrator, MgrModule):
                 self._refresh_host_daemons(host)
         # <service_map>
         sm = {}  # type: Dict[str, orchestrator.ServiceDescription]
-        for h, dm in self.cache.daemons.items():
+        for h, dm in self.cache.get_daemons_with_volatile_status():
             for name, dd in dm.items():
                 if service_type and service_type != dd.daemon_type:
                     continue
@@ -1925,7 +1937,7 @@ class CephadmOrchestrator(orchestrator.Orchestrator, MgrModule):
                 for hostname, hi in self.inventory.items():
                     self._refresh_host_daemons(hostname)
         result = []
-        for h, dm in self.cache.daemons.items():
+        for h, dm in self.cache.get_daemons_with_volatile_status():
             if host and h != host:
                 continue
             for name, dd in dm.items():

diff --git a/src/pybind/mgr/orchestrator/_interface.py b/src/pybind/mgr/orchestrator/_interface.py
index a02572a074e3c3ad7395156b73fd6a52846b80bc..df9b94082202cbecf952558808e5d482503be81b 100644 (file)
--- a/src/pybind/mgr/orchestrator/_interface.py
+++ b/src/pybind/mgr/orchestrator/_interface.py
@@ -1323,6 +1323,10 @@ class DaemonDescription(object):
                 c[k] = datetime.datetime.strptime(c[k], DATEFMT)
         return cls(**c)
 
+    def __copy__(self):
+        # feel free to change this:
+        return DaemonDescription.from_json(self.to_json())
+
 class ServiceDescription(object):
     """
     For responding to queries about the status of a particular service,

diff --git a/src/pybind/mgr/orchestrator/module.py b/src/pybind/mgr/orchestrator/module.py
index d3dbf645fdfad29e37417a2fdd684204d730b7ac..97a8a4962830a5c078acbbba584da8c85902adc5 100644 (file)
--- a/src/pybind/mgr/orchestrator/module.py
+++ b/src/pybind/mgr/orchestrator/module.py
@@ -25,7 +25,7 @@ from ._interface import OrchestratorClientMixin, DeviceLightLoc, _cli_read_comma
     raise_if_exception, _cli_write_command, TrivialReadCompletion, OrchestratorError, \
     NoOrchestrator, OrchestratorValidationError, NFSServiceSpec, \
     RGWSpec, InventoryFilter, InventoryHost, HostSpec, CLICommandMeta, \
-    ServiceDescription, IscsiServiceSpec
+    ServiceDescription, DaemonDescription, IscsiServiceSpec
 
 def nice_delta(now, t, suffix=''):
     if t:
@@ -410,7 +410,7 @@ class OrchestratorCli(OrchestratorClientMixin, MgrModule):
                                        refresh=refresh)
         self._orchestrator_wait([completion])
         raise_if_exception(completion)
-        daemons = completion.result
+        daemons: List[DaemonDescription] = completion.result
 
         def ukn(s):
             return '<unknown>' if s is None else s
@@ -432,12 +432,15 @@ class OrchestratorCli(OrchestratorClientMixin, MgrModule):
             table.left_padding_width = 0
             table.right_padding_width = 2
             for s in sorted(daemons, key=lambda s: s.name()):
-                status = {
-                    -1: 'error',
-                    0: 'stopped',
-                    1: 'running',
-                    None: '<unknown>'
-                }[s.status]
+                if s.status_desc:
+                    status = s.status_desc
+                else:
+                    status = {
+                        -1: 'error',
+                        0: 'stopped',
+                        1: 'running',
+                        None: '<unknown>'
+                    }[s.status]
                 if s.status == 1 and s.started:
                     status += ' (%s)' % to_pretty_timedelta(now - s.started)