From: Adam King <adking@redhat.com>
Date: Fri, 10 Sep 2021 17:09:50 +0000 (-0400)
Subject: mgr/cephadm: delay actions on agent daemons if root cert not created
X-Git-Tag: v17.1.0~816^2~2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=64c2a0326cf114c42e2027c47c6a7377e7c2dbbc;p=ceph.git

mgr/cephadm: delay actions on agent daemons if root cert not created

Signed-off-by: Adam King <adking@redhat.com>
---

diff --git a/src/pybind/mgr/cephadm/agent.py b/src/pybind/mgr/cephadm/agent.py
index a41c1c7f1910..694dd1cff311 100644
--- a/src/pybind/mgr/cephadm/agent.py
+++ b/src/pybind/mgr/cephadm/agent.py
@@ -172,7 +172,7 @@ class HostData:
             else:
                 # we got old counter value with message, inform agent of new timestamp
                 self.mgr.agent_helpers._request_agent_acks({host})
-                self.mgr.log.debug(
+                self.mgr.log.info(
                     f'Received old metadata from agent on host {host}. Requested up-to-date metadata.')
 
             if 'ls' in data and data['ls']:
@@ -187,7 +187,7 @@ class HostData:
 
             if up_to_date:
                 self.mgr.cache.metadata_up_to_date[host] = True
-                self.mgr.log.debug(
+                self.mgr.log.info(
                     f'Received up-to-date metadata from agent on host {host}.')
 
         except Exception as e:
@@ -205,6 +205,7 @@ class AgentMessageThread(threading.Thread):
         super(AgentMessageThread, self).__init__(target=self.run)
 
     def run(self) -> None:
+        self.mgr.log.info(f'Sending message to agent on host {self.host}')
         try:
             assert self.mgr.cherrypy_thread
             root_cert = self.mgr.cherrypy_thread.ssl_certs.get_root_cert()
@@ -250,7 +251,7 @@ class AgentMessageThread(threading.Thread):
                 msg = (bytes_len + self.data)
                 secure_agent_socket.sendall(msg.encode('utf-8'))
                 agent_response = secure_agent_socket.recv(1024).decode()
-                self.mgr.log.debug(f'Received "{agent_response}" from agent on host {self.host}')
+                self.mgr.log.info(f'Received "{agent_response}" from agent on host {self.host}')
                 return
             except ConnectionError as e:
                 # if it's a connection error, possibly try to connect again.
diff --git a/src/pybind/mgr/cephadm/serve.py b/src/pybind/mgr/cephadm/serve.py
index c737e481209d..32787a2c8349 100644
--- a/src/pybind/mgr/cephadm/serve.py
+++ b/src/pybind/mgr/cephadm/serve.py
@@ -672,6 +672,15 @@ class CephadmServe:
             return False
         self.log.debug('Applying service %s spec' % service_name)
 
+        if service_type == 'agent':
+            try:
+                assert self.mgr.cherrypy_thread
+                assert self.mgr.cherrypy_thread.ssl_certs.get_root_cert()
+            except Exception:
+                self.log.info(
+                    'Delaying applying agent spec until cephadm endpoint root cert created')
+                return False
+
         self._apply_service_config(spec)
 
         if service_type == 'osd':
@@ -931,6 +940,15 @@ class CephadmServe:
             if dd.service_name() in self.mgr.spec_store.spec_deleted:
                 continue
 
+            if dd.daemon_type == 'agent':
+                try:
+                    assert self.mgr.cherrypy_thread
+                    assert self.mgr.cherrypy_thread.ssl_certs.get_root_cert()
+                except Exception:
+                    self.log.info(
+                        f'Delaying checking {dd.name()} until cephadm endpoint finished creating root cert')
+                    continue
+
             # These daemon types require additional configs after creation
             if dd.daemon_type in REQUIRES_POST_ACTIONS:
                 daemons_post[dd.daemon_type].append(dd)