Improve patronictl multisite support

Author: ants

patroni/api.py

@@ -680,7 +680,7 @@ def do_GET_metrics(self) -> None:
         self.write_response(200, '\n'.join(metrics)+'\n', content_type='text/plain')
 
     def do_GET_multisite(self):
-        self._write_json_response(200, {"status": self.server.patroni.multisite.status()})
+        self._write_json_response(200, self.server.patroni.multisite.status())
 
     def _read_json_content(self, body_is_optional: bool = False) -> Optional[Dict[Any, Any]]:
         """Read JSON from HTTP request body.
@@ -1401,6 +1401,10 @@ def get_postgresql_status(self, retry: bool = False) -> Dict[str, Any]:
         if self.server.patroni.ha.failsafe_is_active():
             result['failsafe_mode_is_active'] = True
         result['dcs_last_seen'] = self.server.patroni.dcs.last_seen
+
+        if self.server.patroni.multisite.is_active:
+            result['multisite'] = self.server.patroni.multisite.status()
+
         return result
 
     def handle_one_request(self) -> None:

patroni/ctl.py

@@ -1502,16 +1502,14 @@ def get_cluster_service_info(cluster: Dict[str, Any]) -> List[str]:
     """
     service_info: List[str] = []
 
-    leader_name = cluster['leader']['name'] if cluster.get('leader') else None
-    for m in cluster['members']:
-        if m['name'] == leader_name:
-            try:
-                response = request_patroni(m, endpoint="multisite")
-                msdata = json.loads(response.data.decode('utf-8'))
-                if 'status' in msdata:
-                    service_info.append(f"Multisite {msdata['status']}")
-            except Exception:
-                pass
+
+
+    if 'multisite' in cluster:
+        info = f"Multisite {cluster['multisite']['name'] or ''} is {cluster['multisite']['status'].lower()}"
+        standby_config = cluster['multisite'].get('standby_config', {})
+        if standby_config and standby_config.get('host'):
+            info += f", replicating from {standby_config['host']}:{standby_config.get('port', 5432)}"
+        service_info.append(info)
 
     if cluster.get('pause'):
         service_info.append('Maintenance mode: on')

patroni/ha.py

@@ -507,6 +507,9 @@ def touch_member(self) -> bool:
             if self.is_paused():
                 data['pause'] = True
 
+            if self.patroni.multisite.is_active:
+                data['multisite'] = self.patroni.multisite.status()
+
             ret = self.dcs.touch_member(data)
             if ret:
                 new_state = (data['state'], data['role'])

patroni/multisite.py

@@ -6,7 +6,7 @@
 import time
 import six
 
-from .dcs import Member
+from .dcs import Member, Cluster
 from .dcs.kubernetes import catch_kubernetes_errors, Kubernetes
 from .exceptions import DCSError
 
@@ -59,7 +59,7 @@ def on_shutdown(self, checkpoint_location):
 class SingleSiteController(AbstractSiteController):
     """Do nothing controller for single site operation."""
     def status(self):
-        return "Leader"
+        return {"status": "Leader", "active": False}
 
 class MultisiteController(Thread, AbstractSiteController):
     is_active = True
@@ -117,7 +117,12 @@ def __init__(self, config, on_change=None):
         self._dcs_error = None
 
     def status(self):
-        return "Leader" if self._has_leader else "Standby"
+        return {
+            "status": "Leader" if self._has_leader or self._standby_config is None else "Standby",
+            "active": True,
+            "name": self.name,
+            "standby_config": self.get_active_standby_config(),
+        }
 
     def get_active_standby_config(self):
         return self._standby_config
@@ -317,7 +322,7 @@ def _update_history(self, cluster):
             else:
                 self.dcs.set_history_value(json.dumps([{'last_leader': self.name, 'switches': 0}]))
 
-    def _check_for_failover(self, cluster):
+    def _check_for_failover(self, cluster: Cluster):
         if cluster.failover and cluster.failover.target_site:
             if cluster.failover.target_site == self.name:
                 logger.info("Cleaning up failover key targeting us")

patroni/utils.py

@@ -950,9 +950,17 @@ def cluster_as_json(cluster: 'Cluster') -> Dict[str, Any]:
 
     ret: Dict[str, Any] = {'members': []}
     sync_role = 'quorum_standby' if config.is_quorum_commit_mode else 'sync_standby'
+    multisite_active = False
+    multisite_info = {'status': 'leaderless'}
     for m in cluster.members:
+        multisite = m.data.get('multisite', {})
+        multisite_active = multisite_active or multisite.get('active', False)
         if m.name == leader_name:
-            role = 'standby_leader' if config.is_standby_cluster else 'leader'
+            multisite_standby = multisite.get('status') == 'Standby'
+            multisite_info['status'] = multisite.get('status')
+            multisite_info['name'] = multisite.get('name')
+            multisite_info['standby_config'] = multisite.get('standby_config', {})
+            role = 'standby_leader' if config.is_standby_cluster or multisite_standby else 'leader'
         elif config.is_synchronous_mode and cluster.sync.matches(m.name):
             role = sync_role
         else:
@@ -984,6 +992,8 @@ def cluster_as_json(cluster: 'Cluster') -> Dict[str, Any]:
     ret['members'].sort(key=cmp)
     if config.is_paused:
         ret['pause'] = True
+    if multisite_active:
+        ret['multisite'] = multisite_info
     if cluster.failover and cluster.failover.scheduled_at:
         ret['scheduled_switchover'] = {'at': cluster.failover.scheduled_at.isoformat()}
         if cluster.failover.leader: