[bgp] Fix native OVN BGP CI workarounds#4004
Conversation
The gateway IP (e.g. `192.168.133.1`) was configured on the router loopback interface so that VMs could ping the external subnet gateway. This worked with `ovn-bgp-agent`, but with native OVN BGP — which replaces `ovn-bgp-agent` in RHOSO — pinging the gateway IP fails: OVN's `arp_proxy` responds to the ARP request and the ICMP reaches the router, but an anti-loop flow in `lr_in_ip_input` drops the reply because the source IP matches the router port address. Since BGP routing does not depend on this loopback entry, remove it. Related-Issue: #OSPRH-30905 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Eduardo Olivares <eolivare@redhat.com>
After a fresh deployment, the BGP reconciler's `full_sync()` can be skipped if the OVSDB lock is not yet held at startup, and it is never retried. This leaves `arp_proxy` unset on interconnect LSPs. Restarting the neutron pods triggers a new `full_sync()` that completes the setup. This workaround should be removed once the bug is fixed in neutron. Related-Issue: #OSPRH-30900 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Eduardo Olivares <eolivare@redhat.com>
|
Build failed (check pipeline). Post ✔️ openstack-k8s-operators-content-provider SUCCESS in 22m 34s |
|
recheck |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: evallesp The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
openshift-merge-bot
Bot
merged commit 1e9b6ba
into
openstack-k8s-operators:main
4 participants
Summary
192.168.133.1) from therouter loopback in
prepare-bgp-spines-leaves.yaml. This IP wasneeded with
ovn-bgp-agent, but with native OVN BGP the ping replyis dropped by an OVN anti-loop flow in
lr_in_ip_input. BGP routingdoes not depend on this loopback entry.
prepare-bgp-computes.yaml. After a fresh deployment the BGPreconciler's
full_sync()can be skipped if the OVSDB lock is notyet held, leaving
arp_proxyunset on interconnect LSPs. A podrestart triggers a new
full_sync(). This workaround should beremoved once the bug is fixed in neutron.
Related-Issue: #OSPRH-30905
Related-Issue: #OSPRH-30900
Test plan
loopback gateway IP task
OpenStackControlPlanereconciles successfullyarp_proxyis set on interconnect LSPs after the restartAssisted-By: Claude Opus 4.6 (1M context) noreply@anthropic.com