From 44c47421785ba07d4b238bba06eacc42827c84ab Mon Sep 17 00:00:00 2001
From: Balazs Gibizer <balazs.gibizer@est.tech>
Date: Wed, 25 Mar 2020 17:38:14 +0100
Subject: [PATCH] Reproduce bug 1869050

This patch adds a functional test that reproduce the bug when stale
scheduler instance info prevents booting server with anti-affinity.

Change-Id: If485330b48ae2671651aafabc93f92a8999f7ca2
Related-Bug: #1869050
(cherry picked from commit b52c483308f32f3744dd8a5df424b9f518c13155)
(cherry picked from commit 016eeec9841116bbbbc6c3019850c18012e3781a)
(cherry picked from commit 66e4d8218133a5e4a68f68a3017446cb585675c4)
(cherry picked from commit 2a15e0096bc87234a930bb75b73d4874f0f7ec87)
---
 nova/tests/functional/test_server_group.py | 47 ++++++++++++++++++++++
 1 file changed, 47 insertions(+)

diff --git a/nova/tests/functional/test_server_group.py b/nova/tests/functional/test_server_group.py
index e6268e6bc0db..f3e43bc44131 100644
--- a/nova/tests/functional/test_server_group.py
+++ b/nova/tests/functional/test_server_group.py
@@ -369,6 +369,53 @@ class ServerGroupTestV21(ServerGroupTestBase):
         self.assertNotEqual(servers[0]['OS-EXT-SRV-ATTR:host'],
                             migrated_server['OS-EXT-SRV-ATTR:host'])
 
+    def test_migrate_with_anti_affinity_stale_scheduler_instance_info(self):
+        # Start additional host to test migration with anti-affinity
+        compute3 = self.start_service('compute', host='host3')
+
+        # make sure that compute syncing instance info to scheduler
+        # this tells the scheduler that it can expect such updates periodically
+        # and don't have to look into the db for it at the start of each
+        # scheduling
+        for compute in [self.compute, self.compute2, compute3]:
+            compute.manager._sync_scheduler_instance_info(
+                context.get_admin_context())
+
+        created_group = self.api.post_server_groups(self.anti_affinity)
+        servers = self._boot_servers_to_group(created_group)
+
+        post = {'migrate': {}}
+        self.admin_api.post_server_action(servers[1]['id'], post)
+        migrated_server = self._wait_for_state_change(
+            self.admin_api, servers[1], 'VERIFY_RESIZE')
+
+        self.assertNotEqual(servers[0]['OS-EXT-SRV-ATTR:host'],
+                            migrated_server['OS-EXT-SRV-ATTR:host'])
+
+        # We have 3 hosts, so after the move is confirmed one of the hosts
+        # should be considered empty so we could boot a 3rd server on that host
+        post = {'confirmResize': {}}
+        self.admin_api.post_server_action(servers[1]['id'], post)
+        self._wait_for_state_change(self.admin_api, servers[1], 'ACTIVE')
+
+        # NOTE(gibi): This is bug 1869050. The confirm resize does to update
+        # the scheduler instance info so the migrate_server occupies two host
+        # according to the stale information in the scheduler.
+        # Alternatively waiting for the periodic _sync_scheduler_instance_info
+        # call would update the stale data.
+
+        server3 = self._boot_a_server_to_group(
+            created_group, expected_status='ERROR')
+        self.assertIn('No valid host', server3['fault']['message'])
+
+        # When bug 1869050 is fixed the following is expected:
+        # server3 = self._boot_a_server_to_group(created_group)
+        #
+        # # we have 3 servers that should occupy 3 different hosts
+        # hosts = {server['OS-EXT-SRV-ATTR:host']
+        #          for server in [servers[0], migrated_server, server3]}
+        # self.assertEqual(3, len(hosts))
+
     def test_resize_to_same_host_with_anti_affinity(self):
         self.flags(allow_resize_to_same_host=True)
         created_group = self.api.post_server_groups(self.anti_affinity)