[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[HTCondor-users] PRIVATE_NETWORK_INTERFACE ignored for single dual stack interface

Hi all,

this is a follow-up on our dual stack experiments [1]. It might be a bug, but I am not entirely convinced that I am not just overthinking things.

PRIVATE_NETWORK_INTERFACE is ignored even if a node has one address of IPv4 and IPv6 each, but both are on the same interface.

We are currently trying to safely deploy IPv6 on our cluster. The idea is to have condor prefer our established, private IPv4 to keep the cluster working, but allow using whatever is available when talking to the outside.
On our publicly reachable nodes, this is easily possible using a configuration like this:

	PRIVATE_NETWORK_NAME = "MegaPrivateSuperNetworkName"

This will choose the private IPv4 for internal communication, but fall back to public IPv4 otherwise. For example, it allows the Collector to talk to fixed worker nodes privately, and to cloud nodes publicly.
With this setup, we can just add a public IPv6 without anything breaking - all internal nodes still use the private IPv4.

However, this fails if we have only a private IPv4. In this case, condor completely ignores PRIVATE_NETWORK_INTERFACE.
This makes sense as long as there is only the IPv4 address. But if we add a (public) IPv6 to that same interface, condor still ignores PRIVATE_NETWORK_INTERFACE.
Instead, it announces both private IPv4 and the IPv6 address as equal. This causes IPv6 to be picked even for internal communication.

This seems to be tied to the number of network interfaces. If I force using the loopback in addition, PRIVATE_NETWORK_INTERFACE is honoured.
Is there perhaps any way to get condor to always respect PRIVATE_NETWORK_INTERFACE?



Attachment: smime.p7s
Description: S/MIME cryptographic signature