fail to install Calico with the VPP data plane on an EKS cluster with the DPDK driver

[yehodab]

Environment

• Calico/VPP version: 3.30
• Kubernetes version: 1.33
• Deployment type: EKS
• Network configuration: Calico + VPP + DPDK

Issue description
unable to install calico VPP on EKS

To Reproduce
try to follow the steps described in the documenataion to install Install Calico with the VPP data plane on an EKS cluster with the DPDK driver (https://docs.tigera.io/calico/latest/getting-started/kubernetes/vpp/getting-started#install-calico-with-the-vpp-data-plane-on-an-eks-cluster-with-the-dpdk-driver)

Expected behavior
cluster with vpp + DPDK will be created on EKS

Additional context
It seems that the VPP dataplane for Calico on EKS has not been thoroughly tested recently, especially with the transition from Amazon Linux 2 (AL2) to Amazon Linux 2023 (AL2023).

Previously, the default AL2 AMI for EKS was probably tested with Calico VPP. However, AL2 is now deprecated and will soon lose support, with newer Kubernetes versions (e.g., 1.33) already incompatible with it.

Our attempts to use AL2023 have encountered issues. While we successfully updated the YAML to use ens5 instead of eth0 for the interface, the VPP daemonset fails when executing its pre-start hooks. These hooks include commands like sudo systemctl stop network; sudo systemctl kill network, which are no longer functional in AL2023 (they return "Failed to stop network.service: Unit network.service not loaded").

Even after attempting to modify the hook command to sudo systemctl stop systemd-networkd, the pods still fail to start with the error: "Error creating uplink interface ens5: error trying to find interface with tag main-ens5)".

We are seeking assistance from anyone who has recently successfully installed Calico with VPP on EKS, particularly with AL2023, to help us resolve these integration challenges.

[onong]

Hi @yehodab ,

Apologies for the delay and thanks for bringing this to notice. The details you shared in the description was really helpful and made the job easier. Many thanks. Here's what is happening:

Amazon Linux 2 (AL2) has been deprecated and in its place has come Amazon Linux 2023 (AL2023). AL2023 has many improvements and many more changes and sadly it breaks the deployment of Calico/VPP in eks. Calico/VPP causes the uplink interface to vanish from the root network namespace which confuses systemd-networkd and it reacts by wiping out the DNS config and refusing to configure/manage the tuntap interface created in place of the original uplink interface which in turn causes the kernel to purge the tuntap interface at the expiry of the 1 hour DHCP lease.

[onong]

Pls refer to #794 for the fix details.