Kube-router with graceful restart does not advertise external addresses

[rkojedzinszky]

What happened?
Enabled graceful restart feature (--bgp-graceful-restart)

What did you expect to happen?
I expected that when bgp session comes up, service externalIPs get advertised immediately.

How can we reproduce the behavior you experienced?
Steps to reproduce the behavior:

1. Enable --bgp-graceful-restart and --advertise-external-ip
2. Configure an external bgp speaker with graceful restart enabled too, with ipv4 only configuration
3. Create a service with externalIPs addresses
4. Observe that session comes up, but takes 6 minutes to prefixes appear on external speaker

** System Information (please complete the following information):**

• Kube-Router Version: v1.5.4
• Kube-Router Parameters:

       - --run-router=true
       - --run-firewall=true
       - --run-service-proxy=true
       - --advertise-pod-cidr=false
       - --advertise-external-ip
       - --kubeconfig=/var/lib/kube-router/kubeconfig
       - --iptables-sync-period=24h
       - --ipvs-sync-period=24h
       - --routes-sync-period=24h
       - --auto-mtu=false
       - --bgp-graceful-restart

• Kubernetes Version (kubectl version) : v1.26
• Cloud Type: bare metal
• Kubernetes Deployment Type: kubeadm
• Kube-Router Deployment Type: daemonset
• Cluster Size: 10 nodes

Additional context
I still assume that is a bug in GoBGP, already opened an issue there, howewer there's no response. osrg/gobgp#2596. At the end, the final product is kube-router itself which behaves wrong.

[aauren]

@rkojedzinszky - I took a look at the thread again, but still don't see anything that kube-router can do here to fix it.

I'm not willing to maintain a fork of GoBGP. Neither can we revert to the previous functionality because it was wrong and it will cause a break in functionality for other users.

If you have an alternative that doesn't involve reverting to the previous functionality and is within the code base control of kube-router, let me know here and we can talk through the use case to see if it would make sense for you to submit a PR with the functionality.

In the meantime, I pinged the primary maintainer of GoBGP on your issue. I have found him to be responsive when I bubble things up to him directly so we can see if that gets us any forward movement on your issue upstream.

[rkojedzinszky]

@aauren Thanks for handling it. I totally agree, kube-router behaves right, no modification should be done to it. I hope as a GoBGP user we can get more support from GoBGP team to handle this. Unfortunately earlier I was unable to reach attention, my issue has no replies. Thanks for pushing it!

[vasiliev-a-a]

Hello. May be it is too late, but is not the --bgp-graceful-restart-deferral-time controls the time lag between GoBGP starts up and begins to advertise? It defaults 6m, as it stated in the docs.

[rkojedzinszky]

@vasiliev-a-a That was discussed already in #1389 (comment)

[rkojedzinszky]

@aauren do you have updates on this? I know it is out of kube-router's scope, howewer at the end, as a user, I am facing with an issue that kube-router configured to support graceful-restart, and configured with an external router without GR may end up in prefixes not being advertised at all, manual intervention is needed to solve such a case.

Thanks,

[aauren]

Unfortunately, I don't have any updates on it. Its probably best to just keep following along with the upstream bug. Either that or get into the code for GoBGP and submit an PR.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stale for 5 days with no activity.