From 58b632569496961ac9ffb09c00e043627791bb6e Mon Sep 17 00:00:00 2001
From: okozachenko1203 <okozachenko1203@users.noreply.github.com>
Date: Mon, 22 May 2023 19:39:25 +1000
Subject: [PATCH] fix: Wait until observedGeneration of the capi cluster is
 increased in cluster upgrade

fix #57
---
 magnum_cluster_api/driver.py     | 20 ++++++++++++++++++--
 magnum_cluster_api/exceptions.py |  4 ++++
 magnum_cluster_api/resources.py  |  6 ++++++
 3 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/magnum_cluster_api/driver.py b/magnum_cluster_api/driver.py
index 69f473f8..1475b75f 100644
--- a/magnum_cluster_api/driver.py
+++ b/magnum_cluster_api/driver.py
@@ -12,11 +12,13 @@
 # License for the specific language governing permissions and limitations
 # under the License.
 
+import time
+
 import keystoneauth1
 from magnum import objects as magnum_objects
 from magnum.drivers.common import driver
 
-from magnum_cluster_api import clients, monitor, objects, resources, utils
+from magnum_cluster_api import clients, exceptions, monitor, objects, resources, utils
 
 
 class BaseDriver(driver.Driver):
@@ -56,7 +58,6 @@ def update_cluster_status(self, context, cluster, use_admin_ctx=False):
             self.update_nodegroup_status(context, cluster, node_group)
             for node_group in cluster.nodegroups
         ]
-
         # TODO: watch for topology change instead
         osc = clients.get_openstack_api(context)
 
@@ -182,9 +183,23 @@ def upgrade_cluster(
         """
         # TODO: nodegroup?
 
+        # Get current generation
+        current_generation = resources.Cluster(
+            context, self.k8s_api, cluster
+        ).get_observed_generation()
         resources.apply_cluster_from_magnum_cluster(
             context, self.k8s_api, cluster, cluster_template=cluster_template
         )
+        # Wait till the generation has been increased
+        retry_count = 0
+        while retry_count < 10:
+            new_generation = resources.Cluster(
+                context, self.k8s_api, cluster
+            ).get_observed_generation()
+            if current_generation != new_generation:
+                return
+            time.sleep(1)
+        raise exceptions.ClusterAPIReconcileTimeout()
 
     def delete_cluster(self, context, cluster):
         # NOTE(mnaser): This should be removed when this is fixed:
@@ -254,6 +269,7 @@ def update_nodegroup_status(self, context, cluster, nodegroup):
 
     def update_nodegroup(self, context, cluster, nodegroup):
         # TODO
+
         resources.apply_cluster_from_magnum_cluster(context, self.k8s_api, cluster)
 
     def delete_nodegroup(self, context, cluster, nodegroup):
diff --git a/magnum_cluster_api/exceptions.py b/magnum_cluster_api/exceptions.py
index 5e4a33af..cc0c4341 100644
--- a/magnum_cluster_api/exceptions.py
+++ b/magnum_cluster_api/exceptions.py
@@ -43,3 +43,7 @@ class OpenStackClusterNetworkNotReady(OpenStackClusterNotReady):
 
 class OpenStackClusterSubnetNotReady(OpenStackClusterNotReady):
     pass
+
+
+class ClusterAPIReconcileTimeout(Exception):
+    pass
diff --git a/magnum_cluster_api/resources.py b/magnum_cluster_api/resources.py
index f0769e3c..5ea5cde0 100644
--- a/magnum_cluster_api/resources.py
+++ b/magnum_cluster_api/resources.py
@@ -1348,6 +1348,12 @@ def get_or_none(self) -> objects.Cluster:
             name=utils.get_or_generate_cluster_api_name(self.api, self.cluster)
         )
 
+    def get_observed_generation(self) -> int:
+        capi_cluster = self.get_or_none()
+        if capi_cluster:
+            return capi_cluster.obj["status"]["observedGeneration"]
+        raise Exception("Cluster doesn't exists.")
+
     def get_object(self) -> objects.Cluster:
         return objects.Cluster(
             self.api,