Merge "Fix race condition in members batch update API call"

octavia/api/v2/controllers/member.py

@@ -336,7 +336,6 @@ class MembersController(MemberController):
 
         with context.session.begin():
             db_pool = self._get_db_pool(context.session, self.pool_id)
-            old_members = db_pool.members
 
             project_id, provider = self._get_lb_project_id_provider(
                 context.session, db_pool.load_balancer_id)
@@ -354,6 +353,11 @@ class MembersController(MemberController):
         with context.session.begin():
             self._test_lb_and_listener_and_pool_statuses(context.session)
 
+            # Reload the pool, the members may have been updated between the
+            # first query in this function and the lock of the loadbalancer
+            db_pool = self._get_db_pool(context.session, self.pool_id)
+            old_members = db_pool.members
+
             old_member_uniques = {
                 (m.ip_address, m.protocol_port): m.id for m in old_members}
             new_member_uniques = [

releasenotes/notes/fix-race-condition-member-batch-update-1aed0e06004c5dad.yaml

@@ -0,0 +1,7 @@
+---
+fixes:
+  - |
+    Fixed a race condition in the members batch update API call, the data
+    passed to the Octavia worker service may have been incorrect when quickly
+    sending successive API calls. Then the load balancer was stuck in
+    PENDING_UPDATE provisioning_status.