Split View: eBPF 기반 Kubernetes 관측성 가이드: Cilium Hubble부터 Tetragon까지

# Install Cilium CLI
CILIUM_CLI_VERSION=$(curl -s https://raw.githubusercontent.com/cilium/cilium-cli/main/stable.txt)
curl -L --fail --remote-name-all \
  https://github.com/cilium/cilium-cli/releases/download/${CILIUM_CLI_VERSION}/cilium-linux-amd64.tar.gz
sudo tar xzvfC cilium-linux-amd64.tar.gz /usr/local/bin
rm cilium-linux-amd64.tar.gz

# Install Cilium (with Hubble)
cilium install --version 1.16.5 \
  --set hubble.relay.enabled=true \
  --set hubble.ui.enabled=true \
  --set hubble.metrics.enableOpenMetrics=true \
  --set hubble.metrics.enabled="{dns,drop,tcp,flow,port-distribution,icmp,httpV2:exemplars=true;labelsContext=source_ip\,source_namespace\,source_workload\,destination_ip\,destination_namespace\,destination_workload\,traffic_direction}"

# Check status
cilium status
cilium hubble port-forward &

# Observe all network flows
hubble observe

# Flows from a specific namespace only
hubble observe --namespace production

# Observe DNS queries
hubble observe --protocol dns

# Observe HTTP requests (L7)
hubble observe --protocol http

# Observe dropped packets (network policy violations, etc.)
hubble observe --verdict DROPPED

# Communication between specific Pods
hubble observe --from-pod production/frontend --to-pod production/api-server

# JSON output for pipeline integration
hubble observe --namespace production -o json | jq '.flow.source.identity'

# Hubble UI port forwarding
kubectl port-forward -n kube-system svc/hubble-ui 12000:80

# Access http://localhost:12000 in your browser
# View service map + real-time traffic flows

# Prometheus ServiceMonitor
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: hubble-metrics
  namespace: kube-system
spec:
  selector:
    matchLabels:
      k8s-app: hubble
  endpoints:
    - port: hubble-metrics
      interval: 15s

# PromQL queries for Grafana dashboards

# HTTP request rate (by service)
sum(rate(hubble_http_requests_total[5m])) by (destination_workload, method, status)

# DNS error rate
sum(rate(hubble_dns_responses_total{rcode!="No Error"}[5m])) by (source_workload, qtypes)

# Dropped packets (network policy)
sum(rate(hubble_drop_total[5m])) by (reason, source_workload)

# TCP connection time
histogram_quantile(0.99, sum(rate(hubble_tcp_connect_duration_seconds_bucket[5m])) by (le, destination_workload))

# Install with Helm
helm repo add cilium https://helm.cilium.io
helm install tetragon cilium/tetragon -n kube-system

# Install Tetragon CLI
GOOS=$(go env GOOS)
GOARCH=$(go env GOARCH)
curl -L --remote-name-all \
  https://github.com/cilium/tetragon/releases/latest/download/tetra-${GOOS}-${GOARCH}.tar.gz
sudo tar -C /usr/local/bin -xzvf tetra-${GOOS}-${GOARCH}.tar.gz

# All process execution events
kubectl exec -n kube-system ds/tetragon -c tetragon -- \
  tetra getevents -o compact

# Example output:
# 🚀 process default/nginx-7b4f... /bin/sh -c nginx -g 'daemon off;'
# 🚀 process default/nginx-7b4f... /usr/sbin/nginx -g daemon off;
# 💥 exit    default/nginx-7b4f... /bin/sh -c nginx -g 'daemon off;' 0

# Detect sensitive file access
apiVersion: cilium.io/v1alpha1
kind: TracingPolicy
metadata:
  name: sensitive-file-access
spec:
  kprobes:
    - call: 'fd_install'
      syscall: false
      args:
        - index: 0
          type: int
        - index: 1
          type: 'file'
      selectors:
        - matchArgs:
            - index: 1
              operator: 'Prefix'
              values:
                - '/etc/shadow'
                - '/etc/passwd'
                - '/root/.ssh'
                - '/var/run/secrets/kubernetes.io'

# Detect external network connections
apiVersion: cilium.io/v1alpha1
kind: TracingPolicy
metadata:
  name: external-connections
spec:
  kprobes:
    - call: 'tcp_connect'
      syscall: false
      args:
        - index: 0
          type: 'sock'
      selectors:
        - matchArgs:
            - index: 0
              operator: 'NotDAddr'
              values:
                - '10.0.0.0/8'
                - '172.16.0.0/12'
                - '192.168.0.0/16'

# Apply TracingPolicy
kubectl apply -f sensitive-file-access.yaml

# Observe events
kubectl exec -n kube-system ds/tetragon -c tetragon -- \
  tetra getevents -o compact --pods nginx

# Test: attempt to access /etc/shadow from nginx Pod
kubectl exec -it nginx -- cat /etc/shadow
# Tetragon event output:
# 📬 open    default/nginx /etc/shadow

# System call frequency (top 10)
bpftrace -e 'tracepoint:raw_syscalls:sys_enter { @[comm] = count(); }
  interval:s:5 { print(@, 10); clear(@); }'

# Track TCP retransmissions
bpftrace -e 'kprobe:tcp_retransmit_skb {
  @retrans[comm, pid] = count();
}'

# Track file opens
bpftrace -e 'tracepoint:syscalls:sys_enter_openat {
  printf("%s PID:%d -> %s\n", comm, pid, str(args->filename));
}'

# High-latency disk I/O (over 10ms)
bpftrace -e 'kprobe:blk_account_io_done {
  $duration = nsecs - @start[arg0];
  if ($duration > 10000000) {
    printf("slow IO: %d ms, comm=%s\n", $duration/1000000, comm);
  }
}'

#!/usr/bin/env python3
# tcp_latency.py - Track TCP connection latency
from bcc import BPF

bpf_text = """
#include <net/sock.h>
#include <bcc/proto.h>

struct event_t {
    u32 pid;
    u32 daddr;
    u16 dport;
    u64 delta_us;
    char comm[16];
};

BPF_HASH(start, struct sock *);
BPF_PERF_OUTPUT(events);

int trace_connect(struct pt_regs *ctx, struct sock *sk) {
    u64 ts = bpf_ktime_get_ns();
    start.update(&sk, &ts);
    return 0;
}

int trace_tcp_rcv_state_process(struct pt_regs *ctx, struct sock *sk) {
    if (sk->__sk_common.skc_state != TCP_SYN_SENT)
        return 0;

    u64 *tsp = start.lookup(&sk);
    if (!tsp) return 0;

    struct event_t event = {};
    event.pid = bpf_get_current_pid_tgid() >> 32;
    event.delta_us = (bpf_ktime_get_ns() - *tsp) / 1000;
    event.daddr = sk->__sk_common.skc_daddr;
    event.dport = sk->__sk_common.skc_dport;
    bpf_get_current_comm(&event.comm, sizeof(event.comm));

    events.perf_submit(ctx, &event, sizeof(event));
    start.delete(&sk);
    return 0;
}
"""

b = BPF(text=bpf_text)
b.attach_kprobe(event="tcp_v4_connect", fn_name="trace_connect")
b.attach_kprobe(event="tcp_rcv_state_process", fn_name="trace_tcp_rcv_state_process")

def print_event(cpu, data, size):
    event = b["events"].event(data)
    print(f"PID={event.pid} COMM={event.comm.decode()} "
          f"DEST={socket.inet_ntoa(struct.pack('I', event.daddr))}:{socket.ntohs(event.dport)} "
          f"LATENCY={event.delta_us}us")

b["events"].open_perf_buffer(print_event)
while True:
    b.perf_buffer_poll()

Phase 1: Cilium + Hubble (Network Visibility)
  ├── Automatic service map generation
  ├── DNS error detection
  └── Network policy violation monitoring

Phase 2: Tetragon (Security Observability)
  ├── Process execution auditing
  ├── Sensitive file access detection
  └── External connection monitoring

Phase 3: Custom eBPF (Deep Analysis)
  ├── Performance profiling
  ├── Latency analysis
  └── Custom metrics

Cilium Agent: ~200MB RAM, ~0.1 CPU per node
Hubble Relay: ~128MB RAM, ~0.05 CPU
Tetragon: ~256MB RAM, ~0.15 CPU per node

Total overhead: ~1% CPU, ~600MB RAM per node
→ 50-70% savings compared to sidecar proxies (Envoy)