Monitoring Outbound Container Traffic with Tetragon (Without Kubernetes)
How I use Cilium Tetragon, a tiny Docker container-mapper, and Fluent Bit to monitor outbound TCP/UDP connections from my containers on a standalone Docker host—without Kubernetes.
If you run a bunch of Docker containers on a VPS or homeserver, you probably know what's listening on your ports. But do you know what your containers are connecting out to?
I wanted to get visibility into outbound network connections on my standalone Docker host using Cilium Tetragon (eBPF-based security observability). Most Tetragon tutorials assume you're running Kubernetes, where Tetragon automatically enriches events with pod and namespace names. On plain Docker, you only get raw container IDs.
Here's how I solved that with a small three-container docker-compose.yml setup.
The Architecture
My stack in /docker/tetragon consists of three services:
tetragon(quay.io/cilium/tetragon:v1.3.0): Hookstcp_connectandudp_sendmsgin the kernel via eBPF and writesPROCESS_KPROBEevents to a rotating log file.container-mapper(docker:cli): A tiny loop that writesID -> Namemappings fromdocker ps -a --no-truncintocontainers.mapevery 10 seconds.fluent-bit(fluent/fluent-bit:latest): Tailstetragon.log, runs a Lua script (enrich_container.lua) to attach the human-readable container name fromcontainers.map, and forwards the enriched JSON to my log collector.
Filtering Out Noise with a TracingPolicy
By default, tracing every tcp_connect and udp_sendmsg on a host produces a lot of noise (localhost, Docker bridge networks, WireGuard, etc.). In tetragon.tp.d/external-network.yaml, I use a TracingPolicy with NotDAddr selectors to ignore private, loopback, and link-local ranges—plus the server's own public IP:
apiVersion: cilium.io/v1alpha1
kind: TracingPolicy
metadata:
name: external-network-tracing
spec:
kprobes:
- call: "tcp_connect"
syscall: false
args:
- index: 0
type: "sock"
selectors:
- matchArgs:
- index: 0
operator: "NotDAddr"
values:
- "0.0.0.0/8"
- "127.0.0.0/8"
- "10.0.0.0/8"
- "172.16.0.0/12"
- "192.168.0.0/16"
- "169.254.0.0/16"
- "::/128"
- "::1/128"
- "fc00::/7"
- "fe80::/10"
And in the tetragon container command, I restrict the export to PROCESS_KPROBE events only (--export-allowlist={"event_set":["PROCESS_KPROBE"]}), with log rotation enabled (50 MB, 5 backups, compressed).
The Docker Compose Setup
services:
tetragon:
image: quay.io/cilium/tetragon:v1.3.0
container_name: tetragon
restart: unless-stopped
privileged: true
pid: host
network_mode: host
command:
- --tracing-policy-dir=/etc/tetragon/tetragon.tp.d
- --export-filename=/var/run/tetragon/tetragon.log
- --export-file-max-size-mb=50
- --export-file-max-backups=5
- --export-file-compress=true
- --export-allowlist={"event_set":["PROCESS_KPROBE"]}
- --metrics-server=
- --health-server-address=127.0.0.1:6789
volumes:
- /sys/kernel/btf/vmlinux:/var/lib/tetragon/btf:ro
- /docker/tetragon/tetragon.tp.d:/etc/tetragon/tetragon.tp.d:ro
- /docker/tetragon/log:/var/run/tetragon
container-mapper:
image: docker:cli
container_name: tetragon-container-mapper
restart: unless-stopped
entrypoint: ["/bin/sh", "-c"]
command:
- "while true; do docker ps -a --no-trunc --format '\t' > /out/containers.map.tmp && mv /out/containers.map.tmp /out/containers.map; sleep 10; done"
volumes:
- /var/run/docker.sock:/var/run/docker.sock:ro
- /docker/tetragon:/out
fluent-bit:
image: fluent/fluent-bit:latest
container_name: tetragon-fluent-bit
restart: unless-stopped
network_mode: host
depends_on:
- tetragon
- container-mapper
volumes:
- /docker/tetragon/log:/var/run/tetragon:ro
- /docker/tetragon/fluent-bit.conf:/fluent-bit/etc/fluent-bit.conf:ro
- /docker/tetragon/fluent-bit-db:/var/run/fluent-bit
- /docker/tetragon/enrich_container.lua:/fluent-bit/etc/enrich_container.lua:ro
- /docker/tetragon/containers.map:/fluent-bit/etc/containers.map:ro
What I Like (and What to Watch Out For)
- Zero app changes: eBPF runs in the kernel, so none of my containers need sidecars or special flags.
- Readable logs: The
container-mapper+enrich_container.luatrick is dead simple and gives me actual container names instead of 64-character hex IDs. - Privileged mode: Tetragon needs
privileged: true,pid: host, and/sys/kernel/btf/vmlinuxmounted read-only. That's expected for eBPF tooling, but worth keeping in mind.
Overall, it's been running quietly in the background and gives me a clean feed of every external connection leaving the server.