Merge branch 'refresh-v3' into parallel-restart-port

Author: dragomirp

.github/workflows/promote.yaml

@@ -24,8 +24,8 @@ on:
 
 jobs:
   promote:
-    name: Promote charm
-    uses: canonical/data-platform-workflows/.github/workflows/_promote_charm.yaml@v35.0.2
+    name: Promote charms
+    uses: canonical/data-platform-workflows/.github/workflows/_promote_charms.yaml@v39.1.0
     with:
       track: '16'
       from-risk: ${{ inputs.from-risk }}

poetry.lock

@@ -19,7 +19,7 @@ lightkube-models = "^1.28.1.4"
 psycopg2 = "^2.9.10"
 httpx = "^0.28.1"
 postgresql-charms-single-kernel = "16.1.3"
-charm-refresh = "^3.1.0.2"
+charm-refresh = "^3.1.1.2"
 
 [tool.poetry.group.charm-libs.dependencies]
 # data_platform_libs/v0/data_interfaces.py

pyproject.toml

@@ -253,6 +253,8 @@ def __init__(self, *args):
         self.framework.observe(self.on.promote_to_primary_action, self._on_promote_to_primary)
         self.framework.observe(self.on.get_primary_action, self._on_get_primary)
         self.framework.observe(self.on.update_status, self._on_update_status)
+        # Do not use collect status events elsewhere—otherwise ops will prioritize statuses incorrectly
+        # https://canonical-charm-refresh.readthedocs-hosted.com/latest/add-to-charm/status/#implementation
         self.framework.observe(self.on.collect_unit_status, self._reconcile_refresh_status)
         self.framework.observe(self.on.secret_remove, self._on_secret_remove)
 
@@ -278,6 +280,7 @@ def __init__(self, *args):
             except ModelError:
                 logger.exception("failed to open port")
 
+        self.can_set_app_status = True
         try:
             self.refresh = charm_refresh.Kubernetes(
                 PostgreSQLRefresh(
@@ -287,8 +290,10 @@ def __init__(self, *args):
                     _charm=self,
                 )
             )
+        except charm_refresh.KubernetesJujuAppNotTrusted:
+            self.refresh = None
+            self.can_set_app_status = False
         except (
-            charm_refresh.KubernetesJujuAppNotTrusted,
             charm_refresh.PeerRelationNotReady,
             charm_refresh.UnitTearingDown,
         ):
@@ -386,7 +391,9 @@ def _reconcile_refresh_status(self, _=None):
             new_refresh_unit_status = self.refresh.unit_status_higher_priority.message
         elif self.unit.status.message == last_refresh_unit_status:
             if self.refresh is not None and (
-                refresh_status := self.refresh.unit_status_lower_priority()
+                refresh_status := self.refresh.unit_status_lower_priority(
+                    workload_is_running=self._is_workload_running
+                )
             ):
                 self.unit.status = refresh_status
                 new_refresh_unit_status = refresh_status.message
@@ -396,7 +403,11 @@ def _reconcile_refresh_status(self, _=None):
         elif (
             isinstance(self.unit.status, ActiveStatus)
             and self.refresh is not None
-            and (refresh_status := self.refresh.unit_status_lower_priority())
+            and (
+                refresh_status := self.refresh.unit_status_lower_priority(
+                    workload_is_running=self._is_workload_running
+                )
+            )
         ):
             self.unit.status = refresh_status
             new_refresh_unit_status = refresh_status.message
@@ -1158,7 +1169,7 @@ def _on_postgresql_pebble_ready(self, event: WorkloadEvent) -> None:
         # Safeguard against starting while refreshing.
         if self.refresh is None:
             logger.warning("Warning on_postgresql_pebble_ready: Refresh could be in progress")
-        elif self.refresh.in_progress:
+        elif self.refresh.in_progress and not self.refresh.next_unit_allowed_to_refresh:
             logger.debug("Defer on_postgresql_pebble_ready: Refresh in progress")
             event.defer()
             return

src/charm.py

@@ -31,7 +31,7 @@ def is_compatible(
         old_workload_version: str,
         new_workload_version: str,
     ) -> bool:
-        """Checks charm version compatibility."""
+        """Checks charm and workload version compatibility."""
         if not super().is_compatible(
             old_charm_version=old_charm_version,
             new_charm_version=new_charm_version,
@@ -51,9 +51,9 @@ def run_pre_refresh_checks_after_1_unit_refreshed(self) -> None:
         """Implement pre-refresh checks after 1 unit refreshed."""
         logger.debug("Running pre-refresh checks")
         if not self._charm._patroni.are_all_members_ready():
-            raise charm_refresh.PrecheckFailed("Not all members are ready yet.")
+            raise charm_refresh.PrecheckFailed("PostgreSQL is not running on 1+ units")
         if self._charm._patroni.is_creating_backup:
-            raise charm_refresh.PrecheckFailed("A backup is being created.")
+            raise charm_refresh.PrecheckFailed("Backup in progress")
 
     def run_pre_refresh_checks_before_any_units_refreshed(self) -> None:
         """Implement pre-refresh checks before any unit refreshed."""
@@ -64,6 +64,7 @@ def run_pre_refresh_checks_before_any_units_refreshed(self) -> None:
         primary_unit_name = self._charm._patroni.get_primary(unit_name_pattern=True)
         unit_zero_name = f"{self._charm.app.name}/0"
         if primary_unit_name != unit_zero_name:
-            raise charm_refresh.PrecheckFailed(
+            logger.warning(
                 f"Switch primary to {unit_zero_name} to avoid multiple switchovers during refresh."
             )
+            raise charm_refresh.PrecheckFailed(f"Switch primary to {unit_zero_name}")

src/refresh.py

@@ -682,6 +682,8 @@ def _remove_previous_cluster_information(self) -> None:
 
     def set_app_status(self) -> None:
         """Set the app status."""
+        if not self.charm.can_set_app_status:
+            return
         if self.charm.refresh is not None and self.charm.refresh.app_status_higher_priority:
             self.charm.app.status = self.charm.refresh.app_status_higher_priority
             return