Skip to content

How to harden a Kafka container (and is apache/kafka:3.9.0 safe for untrusted workloads?)

A broker sits between every service you run, which is exactly why its container should be one of the tightest. A stock docker run apache/kafka:3.9.0 is closer than most, it already runs as a non-root user, but graded on IronClaw's seven-dimension containment scale it still scores only 63 of 100, grade C (partial). Higher is safer. A couple of runtime flags take the same image to 89 of 100, grade B, one point off an A, and the one dimension it cannot reach is the one a broker needs by definition (clients must connect to it). Here are the exact gaps and fixes from the scan data.

Every number here comes from a read-only docker inspect of apache/kafka:3.9.0, the same data behind its isolation scorecard. No workload is executed. How scoring works →

Where the default configuration leaks

ironctl scan grades seven independent containment boundaries. On a default docker run apache/kafka:3.9.0, three fail or warn (and, notably, non-root already passes):

Dimension Verdict Score What the scan found
Non-root user (uid != 0) ✅ PASS 15/15 runs as appuser (uid != 0)
Dropped capabilities ❌ FAIL 4/20 default capability set retained (CAP_NET_RAW, CAP_MKNOD, and more)
Seccomp profile ✅ PASS 15/15 seccomp profile active
Network isolation / egress ⚠️ WARN 4/15 network=bridge: outbound egress is possible
Read-only root filesystem ❌ FAIL 0/10 root filesystem is writable
No docker.sock exposure ✅ PASS 15/15 no control socket mounted
No shared host namespaces ✅ PASS 10/10 no host PID/IPC/network sharing

Apache's image already did the hardest part, it drops root. The sharpest edges that remain are the retained capabilities and the writable rootfs: a plugin or connector CVE that lands code execution lands it with CAP_NET_RAW and friends, on a process every service in your stack already connects to, and with a filesystem it can rewrite to persist.

Harden it: the exact --fix remediation

ironctl scan my-kafka --fix prints one remediation per failed dimension, then one hardened run. For apache/kafka:3.9.0:

  • --cap-drop=ALL (Dropped capabilities, +16): drop every Linux capability; add back only what the workload provably needs. Kafka needs none of the defaults.
  • --read-only --tmpfs /tmp (Read-only rootfs, +10): make the root filesystem read-only and mount /var/lib/kafka as an explicit writable volume. Removes the persistence surface.
  • Scoped network (Network isolation): --network=none scores the full 15 but is wrong for a broker, clients must be able to connect to it. Any named or bridge network scores 4 of 15 (a WARN, not a fail): a connection path exists. This is the one dimension a broker cannot max out. Contain it anyway: attach a user-defined network scoped to just its producers and consumers, with no default route out, so a compromised broker cannot call arbitrary internet addresses.

No --user flag is needed: the image already runs as appuser.

Before and after

# Before: 63/100, grade C
docker run -d --name kafka apache/kafka:3.9.0

# After: 89/100, grade B (scoped private network for producers and consumers)
docker run -d --name kafka-hardened \
  --cap-drop=ALL \
  --security-opt=no-new-privileges \
  --read-only --tmpfs /tmp \
  -v kafka-data:/var/lib/kafka \
  --network=kafka-internal \
  apache/kafka:3.9.0

Rescan: ironctl scan kafka-hardened reports 89/100 grade B. A 26-point swing with no custom image build, just the right flags. The only dimension still short of full marks is the network (4 of 15), because a broker exists to be connected to; network=none would score the last points but leave nothing able to reach the topics. That is the honest ceiling for a broker, and it is a clear step up from the default C.

Verify it on your own broker

# install (Homebrew)
brew install ironsecco/ironclaw/ironclaw

# grade your running container, then print the fixes
ironctl scan my-kafka
ironctl scan my-kafka --fix

ironctl scan also reads a docker-compose.yml service or a Kubernetes manifest, so you can grade the Kafka in your stack, not just a bare docker run.

Keep going