OVS is stopped first and ovn-controller second in multinode tests.
On exit ovn-controller cleans up after itself: it deletes its own
records from the SB database and its chassis-related state from
the OVS database. With OVS already gone, that cleanup can never
complete, and with INLINE_BR_EX=yes eth1 is an OVS internal port
on br-ex, so stopping OVS also cuts the node off from the SB database
while the delete is in flight, so stop ovn-controller first, the same
order systemd uses, where ovn-controller.service is ordered after OVS.

Fixes: 44d3bf28e4b7 ("multinode: Add support for INLINE_BR_EX configuration of 
multinode.")
Signed-off-by: Alexandra Rukomoinikova <[email protected]>
---
 tests/multinode-macros.at | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/tests/multinode-macros.at b/tests/multinode-macros.at
index be177ed43..13e4676ad 100644
--- a/tests/multinode-macros.at
+++ b/tests/multinode-macros.at
@@ -311,9 +311,11 @@ multinode_setup_controller() {
         eth1_ip=$(m_as $c ip -4 -o addr show eth1 | awk '{print $4}')
     fi
 
-    # Cleanup existing one
-    m_as $c /usr/share/openvswitch/scripts/ovs-ctl stop
+    # Cleanup existing one.  Stop ovn-controller before OVS: on exit it removes
+    # its own records from SB, which needs both the OVS database and the SB
+    # connection (with INLINE_BR_EX=yes the latter goes through eth1 on br-ex).
     m_as $c /usr/share/ovn/scripts/ovn-ctl stop_controller
+    m_as $c /usr/share/openvswitch/scripts/ovs-ctl stop
     m_as $c sh -c "rm -f /etc/openvswitch/*.db"
 
     m_as $c /usr/share/openvswitch/scripts/ovs-ctl start --system-id=$c
-- 
2.48.1

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to