netobserv-ebpf-agent

README

Network Observability eBPF Agent

The Network Observability eBPF Agent allows collecting and aggregating all the ingress and egress flows on a Linux host (required a Kernel 4.18+ with eBPF enabled).

• How to compile
• Hot to configure
• How to run
• Development receipts
• Known issues
• Frequently-asked questions
• Troubleshooting

How to compile

make build

To build the agent image and push it to your Docker / Quay repository, run:

IMG=quay.io/myaccount/netobserv-ebpf-agent:dev make image-build image-push

How to configure

The eBPF Agent is configured by means of environment variables. Check the configuration documentation for more details.

How to run

The NetObserv eBPF Agent is designed to run as a DaemonSet in OpenShift/K8s. It is triggered and configured by our Network Observability Operator.

Anyway you can run it directly as an executable from your command line:

export FLOWS_TARGET_HOST=...
export FLOWS_TARGET_PORT=...
sudo -E bin/netobserv-ebpf-agent

To deploy locally, use instructions from flowlogs-dump (like tcpdump).
To deploy it as a Pod, you can check the deployment examples.

The Agent needs to be executed either with:

1. The following Linux capabilities (recommended way): BPF, PERFMON, NET_ADMIN, SYS_RESOURCE. If you deploy it in Kubernetes or OpenShift, the container running the Agent needs to define the following securityContext:

   securityContext:
     runAsUser: 0
     capabilities:
       add:
         - BPF
         - PERFMON
         - NET_ADMIN
         - SYS_RESOURCE
   

   (Please notice that the runAsUser: 0 is still needed).
2. Administrative privileges. If you deploy it in Kubernetes or OpenShift, the container running the Agent needs to define the following securityContext:

   securityContext:
     privileged: true
     runAsUser: 0
   

   This option is only recommended if your Kernel does not recognize some of the above capabilities. We found some Kubernetes distributions (e.g. K3s) that do not recognize the BPF and PERFMON capabilities.

Here is a list of distributions where we tested both full privileges and capability approaches, and whether they worked (✅) or did not (❌):

Distribution	K8s Server version	Capabilities	Privileged
Amazon EKS (Bottlerocket AMI)	1.22.6	✅	✅
K3s (Rancher Desktop)	1.23.5	❌	✅
Kind	1.23.5	❌	✅
OpenShift	1.23.3	✅	✅

Development receipts

How to regenerate the eBPF Kernel binaries

The eBPF program is embedded into the pkg/ebpf/bpf_* generated files. This step is generally not needed unless you change the C code in the bpf folder.

If you have Docker installed, you just need to run:

make docker-generate

If you can't install docker, you should locally install the following required packages:

dnf install -y kernel-devel make llvm clang glibc-devel.i686
make generate

Tested in Fedora 35 and Red Hat Enterprise Linux 8.

Known issues

Extrenal Traffic in Openshift (OVN-Kubernetes CNI)

For egress traffic, you can see the source Pod metadata. For ingress traffic (e.g. an HTTP response), you see the destination Host metadata.

Frequently-asked questions

Where is the collector?

As part of our Network Observability solution, the eBPF Agent is designed to send the traced flows to our Flowlogs Pipeline component.

In addition, we provide a simple GRPC+Protobuf library to allow implementing your own collector. Check the packet counter code for an example of a simple collector using our library.

Troubleshooting

Deployed as a Kubernetes Pod, the agent shows permission errors in the logs and can't start

In your deployment file, make sure that the container runs as the root user (runAsUser: 0) and with the granted capabilities or privileges (see how to run section).

The Agent doesn't work in my Amazon EKS puzzle

Despite Amazon Linux 2 enables eBPF by default in EC2, the EKS images are shipped with disabled eBPF.

You'd need either:

1. Provide your own AMI configured to work with eBPF
2. Use other Linux distributions that are shipped with eBPF enabled by default. We have successfully tested the eBPF Agent in EKS with the Bottlerocket Linux distribution, without requiring any extra configuration.