Hello Sowmini,
Good point! We addressed this in a previous version of the draft, but
accidentally dropped it from the current version.
Our current thinking is:
- If the destination and probed interfaces are in the same VRF, the ICMP
Extended Echo Reply message will reflect the state of the probed interface
- If the destination interface is in the general or management VRF and the
probed interface is in another VRF, the ICMP Extended Echo Reply message will
reflect the state of the probed interface
- Otherwise, the ICMP Extended Echo message will contain an error code
indicating that the probed interface does not exist.
I will add this back to the next revision of the draft.
Ron
> -----Original Message-----
> From: Sowmini Varadhan [mailto:[email protected]]
> Sent: Tuesday, March 28, 2017 5:32 PM
> To: Ron Bonica <[email protected]>; Reji Thomas
> <[email protected]>; [email protected]; [email protected]
> Cc: [email protected]; [email protected]
> Subject: question about xping (draft-bonica-intarea-eping)
>
> Hi,
>
> The draft says
> ".. the destination and probed
> interfaces must be local to one another (i.e., both interfaces must
> belong to the same node)."
>
> However, virtualization raises some questions about how this should work,
> and maybe the draft should nail down some of the details with more clarity.
>
> What if the probed interface is assigned to a different VRF than the
> destination interface? e.g, What is the expected xping response if I receive
> # xping-I red0/0/0.0 10.10.10.1
> on interface blue0 in the "blue" vrf, and red0 is in "red" VRF?
> Does the answer change if the destination interface is on the management
> VRF?
>
> similar question for other forms of virtualization, e.g,, What if the probed
> interface is assigned to a different network-namespace than the destination
> interface of an xping req?
> Does it matter if the above xping is received on the default network
> namespace vs some other network namespace?
>
> One could argue that all the physical intefaces are owned by the "node"
> (the management VRF? default netns?) but virtual interfaces like tunnels and
> macvlans are a grayer area (unlike physical interfaces, the latter do not
> default back to the mgmt VRF or default netns when the virtual object is
> destroyed, so their ownership is unclear).
>
> How should xping work in the face of virtual interfaces, e.g., the linux
> macvlan? How does it work when an interface is renamed?
> Will xping allow me to probe for "tun0" or "Portchannel1"?
>
> --Sowmini
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area
