Remove reinits (#999)

Author: dragomirp

src/charm.py

@@ -913,23 +913,6 @@ def _on_peer_relation_changed(self, event: HookEvent):
             event.defer()
             return
 
-        # Restart the workload if it's stuck on the starting state after a timeline divergence
-        # due to a backup that was restored.
-        if (
-            not self.is_primary
-            and not self.is_standby_leader
-            and (
-                self._patroni.member_replication_lag == "unknown"
-                or int(self._patroni.member_replication_lag) > 1000
-            )
-        ):
-            logger.warning("Degraded member detected: reinitialising unit")
-            self.set_unit_status(MaintenanceStatus("reinitialising replica"))
-            self._patroni.reinitialize_postgresql()
-            logger.debug("Deferring on_peer_relation_changed: reinitialising replica")
-            event.defer()
-            return
-
         self._start_stop_pgbackrest_service(event)
 
         # This is intended to be executed only when leader is reinitializing S3 connection due to the leader change.
@@ -1933,7 +1916,8 @@ def _on_update_status(self, _) -> None:
         if self.primary_endpoint:
             self._update_relation_endpoints()
 
-        if self._handle_workload_failures():
+        if not self._patroni.member_started and self._patroni.is_member_isolated:
+            self._patroni.restart_patroni()
             return
 
         # Update the sync-standby endpoint in the async replication data.
@@ -2054,40 +2038,6 @@ def _handle_processes_failures(self) -> bool:
 
         return False
 
-    def _handle_workload_failures(self) -> bool:
-        """Handle workload (Patroni or PostgreSQL) failures.
-
-        Returns:
-            a bool indicating whether the charm performed any action.
-        """
-        # Restart the workload if it's stuck on the starting state after a restart.
-        try:
-            is_primary = self.is_primary
-            is_standby_leader = self.is_standby_leader
-        except RetryError:
-            return False
-
-        if (
-            not self.has_raft_keys()
-            and not is_primary
-            and not is_standby_leader
-            and not self._patroni.member_started
-            and "postgresql_restarted" in self._peers.data[self.unit]
-            and self._patroni.member_replication_lag == "unknown"
-        ):
-            logger.warning("Workload failure detected. Reinitialising unit.")
-            self.set_unit_status(MaintenanceStatus("reinitialising replica"))
-            self._patroni.reinitialize_postgresql()
-            return True
-
-        # Restart the service if the current cluster member is isolated from the cluster
-        # (stuck with the "awaiting for member to start" message).
-        if not self._patroni.member_started and self._patroni.is_member_isolated:
-            self._patroni.restart_patroni()
-            return True
-
-        return False
-
     def _set_primary_status_message(self) -> None:
         """Display 'Primary' in the unit status message if the current unit is the primary."""
         try:

tests/unit/test_charm.py

@@ -803,7 +803,6 @@ def test_on_update_status(harness):
         ) as _set_primary_status_message,
         patch("charm.Patroni.restart_patroni") as _restart_patroni,
         patch("charm.Patroni.is_member_isolated") as _is_member_isolated,
-        patch("charm.Patroni.reinitialize_postgresql") as _reinitialize_postgresql,
         patch(
             "charm.Patroni.member_replication_lag", new_callable=PropertyMock
         ) as _member_replication_lag,
@@ -874,24 +873,9 @@ def test_on_update_status(harness):
         harness.charm.on.update_status.emit()
         _set_primary_status_message.assert_called_once()
 
-        # Test the reinitialisation of the replica when its lag is unknown
-        # after a restart.
+        # Test call to restart when the member is isolated from the cluster.
         _set_primary_status_message.reset_mock()
-        _is_primary.return_value = False
-        _is_standby_leader.return_value = False
         _member_started.return_value = False
-        _is_member_isolated.return_value = False
-        _member_replication_lag.return_value = "unknown"
-        with harness.hooks_disabled():
-            harness.update_relation_data(
-                rel_id, harness.charm.unit.name, {"postgresql_restarted": "True"}
-            )
-        harness.charm.on.update_status.emit()
-        _reinitialize_postgresql.assert_called_once()
-        _restart_patroni.assert_not_called()
-        _set_primary_status_message.assert_not_called()
-
-        # Test call to restart when the member is isolated from the cluster.
         _is_member_isolated.return_value = True
         with harness.hooks_disabled():
             harness.update_relation_data(
@@ -908,9 +892,6 @@ def test_on_update_status_after_restore_operation(harness):
         patch(
             "charm.PostgresqlOperatorCharm._set_primary_status_message"
         ) as _set_primary_status_message,
-        patch(
-            "charm.PostgresqlOperatorCharm._handle_workload_failures"
-        ) as _handle_workload_failures,
         patch(
             "charm.PostgresqlOperatorCharm._update_relation_endpoints"
         ) as _update_relation_endpoints,
@@ -946,7 +927,6 @@ def test_on_update_status_after_restore_operation(harness):
         _handle_processes_failures.assert_not_called()
         _oversee_users.assert_not_called()
         _update_relation_endpoints.assert_not_called()
-        _handle_workload_failures.assert_not_called()
         _set_primary_status_message.assert_not_called()
         assert isinstance(harness.charm.unit.status, BlockedStatus)
 
@@ -959,7 +939,6 @@ def test_on_update_status_after_restore_operation(harness):
         _handle_processes_failures.assert_not_called()
         _oversee_users.assert_not_called()
         _update_relation_endpoints.assert_not_called()
-        _handle_workload_failures.assert_not_called()
         _set_primary_status_message.assert_not_called()
         assert isinstance(harness.charm.unit.status, ActiveStatus)
 
@@ -973,13 +952,11 @@ def test_on_update_status_after_restore_operation(harness):
         _member_started.return_value = True
         _can_use_s3_repository.return_value = (True, None)
         _handle_processes_failures.return_value = False
-        _handle_workload_failures.return_value = False
         harness.charm.on.update_status.emit()
         _update_config.assert_called_once()
         _handle_processes_failures.assert_called_once()
         _oversee_users.assert_called_once()
         _update_relation_endpoints.assert_called_once()
-        _handle_workload_failures.assert_called_once()
         _set_primary_status_message.assert_called_once()
         assert isinstance(harness.charm.unit.status, ActiveStatus)