Skip to content

Latest commit

 

History

History
358 lines (280 loc) · 9.55 KB

DEVELOPMENT.md

File metadata and controls

358 lines (280 loc) · 9.55 KB
Raw

Develop ebpf locally

First setup lima. It uses nix and devbox to manage most of the packages.

ln -s ~/c/kvisor /tmp/kvisor
limactl start ./tools/lima-ebpf.yaml
limactl shell lima-ebpf
cd /tmp/kvisor
devbox install

Now inside vm you can run ebpftracer

cd /tmp/kvisor/pkg/ebpftracer
go generate ./...
go test -v . -exec sudo -run=TestTracer

Trigger ebpf events

limactl shell lima-ebpf
curl google.com

Run E2E tests locally

You can run tests on your local kind cluster.

KIND_CONTEXT=tilt IMAGE_TAG=local ./e2e/run.sh

Colima

Colima is a wrapper around Lima for macos and can be used as docker desktop for mac replacement.

colima start  --cpu 2 --memory 4 --disk 100 -t vz --mount-type virtiofs

Lima is lower level VM which allows to create customizable templates. This is recommended if you need to work with ebpf code.

Lima

ln -s ~/c/kvisor /tmp/kvisor
limactl start ./tools/lima-ebpf.yaml

TILT local development

1. Install docker, you need 6+ kernel with btf support.

2. Setup local k8s.

You can use kind or any other local k8s cluster. Kind is recommended.

kind create cluster
kubectl cluster-info --context kind-kind

3. Start tilt

tilt up

4. Port-forward server api

kubectl port-forward svc/kvisord-server 6060:80 -n kvisord

5. Run dashboard UI locally

cd ui
npm install
PUBLIC_API_BASE_URL=http://localhost:6060 npm run dev

GKE Cluster

Create cluster

export GCP_PROJECT="my-project"
export CLUSTER_NAME="my-cluster-name"
gcloud beta container --project $GCP_PROJECT \
  clusters create $CLUSTER_NAME \
  --zone "us-central1-c" \
  --cluster-version "1.25.8-gke.500" \
  --machine-type "e2-small" \
  --disk-type "pd-balanced" \
  --disk-size "100" \
  --num-nodes "2" \
  --node-locations "us-central1-c"

Connect

gcloud container clusters get-credentials $CLUSTER_NAME --zone us-central1-c --project $GCP_PROJECT

eBPF

Mount tracepoint events

Mount tracepoints if they are not mounted yet.

mount -t debugfs none /sys/kernel/debug
ls /sys/kernel/debug/tracing/events

Print logs in ebpf

bpf_printk("called");
cat /sys/kernel/debug/tracing/trace_pipe

Clickhouse

Clickhouse show columns data size

select column, formatReadableSize(sum(column_bytes_on_disk)) bytes_on_disk, formatReadableSize(sum(column_data_uncompressed_bytes)) uncompressed
from system.parts_columns
where active = 1 and table like '%events%'
group by database,table, column
order by sum(column_bytes_on_disk) desc;

Query container resource usage stats:

select toStartOfMinute(ts) t,
       case when group = 'cpu' then toString(avg(value)) else formatReadableSize(avg(value)) end as val,
       group||'_'||subgroup name from container_stats
where container_name='kvisor' and group!='syscall' and t > now() - interval 5 minute
group by t, group, name
order by t;

Push chart to gcp artifact registry

Create package and push chart

helm package ./charts/kvisord
helm push kvisord-0.1.6.tgz oci://us-east4-docker.pkg.dev/kvisor/helm-charts

Test chart template

helm template kvisord oci://us-east4-docker.pkg.dev/kvisor/helm-charts/kvisord --version 0.1.6

Public demo

Install kvisord

helm upgrade --install kvisord oci://us-east4-docker.pkg.dev/kvisor/helm-charts/kvisord \
    --version 0.7.0 \
    --namespace kvisord --create-namespace \
    --set storage.resources.requests.cpu=2 \
    --set storage.resources.requests.memory=8Gi \
    --set agent.resources.requests.cpu=100m \
    --set agent.resources.requests.memory=128Mi \
    --set server.resources.requests.cpu=2 \
    --set server.resources.requests.memory=4Gi \
    --set server.extraArgs.events-batch-size=1000 \
    --set server.extraArgs.events-batch-queue-size=30000 \
    --set-string server.extraArgs.workload-profiles-enabled=false \

Open port-forward to local dashboard

kubectl port-forward svc/kvisord-server 6060:80 -n kvisord

Delete kvisord

helm uninstall kvisord -n kvisord

Integrate with CASTAI

helm upgrade --install castai-kvisor oci://us-east4-docker.pkg.dev/kvisor/helm-charts/castai-kvisor \
    --version 0.12.0 \
    --namespace kvisord --create-namespace \
    --set image.tag=0f07db05bbe55f9aba04952337f7023f3a4553e5 \
    --set castai.clusterID=<CLUSTER-ID> \
    --set castai.apiKey=<API-KEY> \
    --set agent.enabled=true \
    --set controller.extraArgs.image-scan-enabled=true \
    --set controller.extraArgs.kube-bench-enabled=true

Misc

List available trace functions for ftrace.

cat /sys/kernel/debug/tracing/available_filter_functions | grep socket_connect

Making new release

  1. Go to https://github.com/castai/kvisor/releases
  2. Click draw new release (should open https://github.com/castai/kvisor/releases/new)
  3. Choose tag. Add new tag. Follow semver. For fixes only bump patch version.
  4. Click generate release notes.
  5. Publish release.

Testing netflow

Install kvisor with netflow export to local clickhouse.

helm repo add castai-helm https://castai.github.io/helm-charts
helm repo update castai-helm

helm upgrade --install castai-kvisor castai-helm/castai-kvisor \
    --namespace castai-agent --create-namespace \
    --set castai.enabled=false \
    --set agent.enabled=true \
    --set agent.extraArgs.netflow-enabled=true \
    --set clickhouse.enabled=true

Check pods are running

kubectl get pods -n castai-agent

You should see agent, clickhouse and controller pods

NAME                                        READY   STATUS    RESTARTS   AGE
castai-kvisor-agent-djjcq                   1/1     Running   0          67s
castai-kvisor-clickhouse-0                  2/2     Running   0          66s
castai-kvisor-controller-8697bbf8cd-sq6jp   1/1     Running   0          67s

Query flows

Port forward clickhouse connection

kubectl port-forward -n castai-agent svc/castai-kvisor-clickhouse 8123

Connect to clickhouse with your favorite sql client with credentials:

Username: kvisor
Password: kvisor
Database: kvisor

Example query:

select toStartOfInterval(ts, INTERVAL 1 HOUR) AS period,
       pod_name,
       namespace,
       workload_name,
       workload_kind,
       zone,
       dst_pod_name,
       dst_namespace,
       dst_domain,
       dst_workload_name,
       dst_workload_kind,
       dst_zone,
       formatReadableSize(sum(tx_bytes)) total_egress,
       formatReadableSize(sum(rx_bytes)) total_ingress
from netflows
group by period,
         pod_name,
         namespace,
         workload_name,
         workload_kind,
         zone,
         dst_pod_name,
         dst_namespace,
         dst_domain,
         dst_workload_name,
         dst_workload_kind,
         dst_zone
order by period;

Generating connections

Start server

docker run -it --rm -p 8000:8000 busybox nc -lk -p 8000

Send messages

for i in {1..50}; do echo "hello$i" | nc -q 0 localhost 8000 & done

Or send messages from single process

cd ./pkg/ebpftracer
NC_ADDR=localhost:8000 go test -v -count=1 . -run=TestGenerateConn

Troubleshooting packets

PWRU

Install pwru

wget https://github.com/cilium/pwru/releases/download/v1.0.6/pwru-linux-amd64.tar.gz && \
tar -xzvf pwru-linux-amd64.tar.gz && \
mv pwru /usr/bin/ && \
rm pwru-linux-amd64.tar.gz

Or with docker

docker run --privileged --rm -t --pid=host -v /sys/kernel/debug/:/sys/kernel/debug/ cilium/pwru pwru --output-tuple 'host 1.1.1.1'

Tracee

docker run --name tracee -it --rm   --pid=host --cgroupns=host --privileged   -v /etc/os-release:/etc/os-release-host:ro   -v /var/run:/var/run:ro   aquasec/tracee:latest --events net_packet_tcp

Linux capabilities

Use inspektor-gadget CLI tool to get the required linux capabilities for the kvisor agent container:

KVISOR_AGENT_KUBERNETES_NODE="kvisor-e2e-control-plane" # Replace with the kvisor agent pod name
kubectl debug --profile=sysadmin node/${KVISOR_AGENT_KUBERNETES_NODE} -it --image=ghcr.io/inspektor-gadget/ig -- \
    ig trace capabilities -c kvisor --unique -t 0

RUNTIME.CONTAINERNAME          PID              COMM             SYSCALL            CAP CAPNAME            AUDIT            VERDICT
kvisor                         69801            kvisor-agent     bpf                39  BPF                1                Allow
kvisor                         69801            kvisor-agent     bpf                38  PERFMON            1                Allow
kvisor                         69801            kvisor-agent     openat             19  SYS_PTRACE         1                Allow
kvisor                         69801            kvisor-agent     readlinkat         19  SYS_PTRACE         1                Allow
kvisor                         69801            kvisor-agent     read               19  SYS_PTRACE         1                Allow
kvisor                         69801            kvisor-agent     setns              21  SYS_ADMIN          1                Allow
kvisor                         69801            kvisor-agent     newfstatat         19  SYS_PTRACE         1                Allow
kvisor                         69801            kvisor-agent     bpf                12  NET_ADMIN          1                Allow
kvisor                         69801            kvisor-agent     perf_event_open    38  PERFMON            1                Allow

Any Deny in the VERDICT column should be added to the containerSecurityContext of the kvisor agent container in the charts/kvisor/values.yaml file to be allowed.