no resilience on network issues

[FaKod]

UseCase:
Running an Exoip Pod in a Kubernetes Cluster. All exoip Container default priority.

exoip removes peer Nics if it fails to connect the node. If the node that could not be connected (in case of a network issue or ACL misconfiguration) is still alive, no one re-adds the Nic again.

[FaKod]

We are currently testing this simple solution: https://github.com/FaKod/exoip/blob/cs_wip_make_it_work/src/exoip/state.go#L20

It calles ObtainNic every 30s (configurable)

[FaKod]

OK, since this runs every 30 seconds on 7 Nodes and 4 Clusters, this leads to some traffic on the Exoscale API.

Thats why I'm testing this now: https://github.com/FaKod/exoip-k8s

It uses Kubernetes Simple Leader Election... this may be a solution.

[marcaurele]

That's still not acceptable for the API calls. You should get the VM's nic and only add the EIP if it's missing.

[greut]

digging old issues here.

Having a master node unsubscribing other ones seems okay. If they cannot communicate directly, you've got another issue. The addIpToNic in a while loop feels wrong.

[greut]

We've been experiencing this kind of behaviour that looks alike:

• two nodes are connected
• they somehow disconnect each other from the eip
• two nodes are present, ping each other, but are disconnected from the eip
• 😿