Skip to content

Multicast discovery issue in Kubernetes #144

Open

Description

@jrz1977

In kubernetes and other cloud providers, multicast is disabled and usually not possible to enable. IPFIX template discovery in this scenario will fail. Is there an alternative method or plan to implement some kind of unicast discovery protocol?

Activity

  1. Slepwin commented on Jan 28, 2021

    @Slepwin

    In kubernetes and other cloud providers, multicast is disabled and usually not possible to enable. IPFIX template discovery in this scenario will fail. Is there an alternative method or plan to implement some kind of unicast discovery protocol?

    In k8s you can use different load-balancing algorithms for service, example for kube-router CNI:

    annotations:
       kube-router.io/service.scheduler: "sh" 
    

    All traffic from one device/source will always going to the same pod and you do not need to use template discovery at all.

  2. jrz1977 commented on Jan 28, 2021

    @jrz1977
    Author

    This could be a

    In kubernetes and other cloud providers, multicast is disabled and usually not possible to enable. IPFIX template discovery in this scenario will fail. Is there an alternative method or plan to implement some kind of unicast discovery protocol?

    In k8s you can use different load-balancing algorithms for service, example for kube-router CNI:

    annotations:
       kube-router.io/service.scheduler: "sh" 
    

    All traffic from one device/source will always going to the same pod and you do not need to use template discovery at all.

    This could be a workaround, but not ideal. Pods are ephemeral by nature and when a pod is recreated, until the device sends the template again, the new pod will not be able to process it.

    Also sticky traffic is not ideal in enterprise networks, for example there could be 1000s of Wifi routers exporting ipfix.

  3. Slepwin commented on Feb 1, 2021

    @Slepwin

    This could be a

    In kubernetes and other cloud providers, multicast is disabled and usually not possible to enable. IPFIX template discovery in this scenario will fail. Is there an alternative method or plan to implement some kind of unicast discovery protocol?

    In k8s you can use different load-balancing algorithms for service, example for kube-router CNI:

    annotations:
       kube-router.io/service.scheduler: "sh" 
    

    All traffic from one device/source will always going to the same pod and you do not need to use template discovery at all.

    This could be a workaround, but not ideal. Pods are ephemeral by nature and when a pod is recreated, until the device sends the template again, the new pod will not be able to process it.

    Also sticky traffic is not ideal in enterprise networks, for example there could be 1000s of Wifi routers exporting ipfix.

    Yes, after pod re-creation need some time (2-3 minutes) to get templates, that not a big problem, also if you have multiple device (routers, etc) you still can balance traffic between pods (with source hash you just sticky traffic from one router to some pod). Sticky traffic is a must for consistent hash.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions