fix(neutron): Fix for neutron+ovn chassis register deleted issue

Author: nicholaskuechler

containers/neutron/Dockerfile

@@ -29,6 +29,16 @@ RUN --mount=type=cache,target=/root/.cache/uv \
         /src/neutron-lib \
         /src/neutron-understack
 
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        patch \
+        quilt \
+    && apt-get clean && rm -rf /var/lib/apt/lists/*
+
+COPY containers/neutron/patches /tmp/patches/
+RUN cd /var/lib/openstack/lib/python3.12/site-packages && \
+    QUILT_PATCHES=/tmp/patches quilt push -a
+
 ARG OPENSTACK_VERSION="required_argument"
 FROM quay.io/airshipit/neutron:${OPENSTACK_VERSION}-ubuntu_noble AS final
 

containers/neutron/patches/fix-ovn-ghost-chassis-agent.patch

@@ -0,0 +1,56 @@
+Description: Prevent ghost OVN agents from orphaned Chassis_Private rows
+ When a Chassis row is deleted but its Chassis_Private persists (e.g.
+ ungraceful ovn-controller shutdown in containers), the ROW_CREATE path
+ in ChassisAgentWriteEvent and ChassisOVNAgentWriteEvent would add an
+ agent to the cache with the DeletedChassis sentinel hostname
+ '("Chassis" register deleted)'. This patch gates ROW_CREATE on
+ row.chassis being non-empty, consistent with the ROW_UPDATE path and
+ the AgentCache.populate() method which already skips such rows.
+Author: Nicholas Kuechler
+
+--- a/neutron/plugins/ml2/drivers/ovn/mech_driver/ovsdb/ovsdb_monitor.py
++++ b/neutron/plugins/ml2/drivers/ovn/mech_driver/ovsdb/ovsdb_monitor.py
+@@ -359,13 +359,17 @@
+ class ChassisAgentWriteEvent(ChassisAgentEvent):
+     events = (BaseEvent.ROW_CREATE, BaseEvent.ROW_UPDATE)
+ 
+     def match_fn(self, event, row, old=None):
+         # On updates to Chassis_Private because the Chassis has been deleted,
+         # don't update the AgentCache. We use chassis_private.chassis to return
+-        # data about the agent.
+-        return (event == self.ROW_CREATE or
+-                (hasattr(old, 'nb_cfg') and row.chassis))
++        # data about the agent. Require a valid chassis reference for all
++        # events including ROW_CREATE: an orphaned Chassis_Private (where the
++        # Chassis row was already deleted) must not create a ghost agent entry.
++        if not row.chassis:
++            return False
++        return (event == self.ROW_CREATE or
++                (hasattr(old, 'nb_cfg') and row.chassis))
+ 
+     def run(self, event, row, old):
+         n_agent.AgentCache().update(ovn_const.OVN_CONTROLLER_AGENT, row,
+                                     clear_down=event == self.ROW_CREATE)
+@@ -427,14 +431,15 @@
+     def match_fn(self, event, row, old=None):
+         if not self.agent_id(row):
+             # Don't create a cached object with an agent_id of 'None'
+             return False
+-        if event == self.ROW_CREATE:
+-            return True
+ 
+-        # On updates to Chassis_Private because the Chassis has been
+-        # deleted, don't update the AgentCache. We use
+-        # chassis_private.chassis to return data about the agent.
++        # Require a valid chassis reference for all events including
++        # ROW_CREATE: an orphaned Chassis_Private (where the Chassis row
++        # was already deleted) must not create a ghost agent entry.
+         if not getattr(row, 'chassis', None):
+             return False
+ 
++        if event == self.ROW_CREATE:
++            return True
++
+         # Check if both rows have external_ids before comparing nb_cfg
+         if not (hasattr(old, 'external_ids') and row.external_ids):
+             return False

containers/neutron/patches/series

@@ -0,0 +1 @@
+fix-ovn-ghost-chassis-agent.patch