Skip to content

hazelcast:5.5.0 container isolation score: 63/100 (grade C)

Run with plain docker run hazelcast/hazelcast:5.5.0 defaults, no hardening flags, the hazelcast image scores 63/100, grade C (partial) on IronClaw's seven-dimension container containment scale. Higher is safer. This is what you get straight out of a copy-pasted docker run; the fixes below show where the lost points are.

Graded from a read-only inspect of a running container started from hazelcast/hazelcast:5.5.0 at digest sha256:5dd5d31c7a0685d832f02eba03556898689ce85796c6b030132e93b7c0d77bd9 with plain docker run defaults, its entrypoint overridden with sleep purely to keep it alive. The scan itself executes nothing inside the container. Scoring an image reference instead of a running container yields a different, non-comparable result. How scoring works →

How it scores, dimension by dimension

Dimension Verdict Score What the scan found
Non-root user (uid != 0) ✅ PASS 15/15 runs as hazelcast (uid != 0)
Dropped capabilities ❌ FAIL 4/20 default capability set retained (includes CAP_NET_RAW, CAP_MKNOD, …)
Seccomp profile ✅ PASS 15/15 seccomp profile active (syscall surface filtered)
Network isolation / egress ⚠️ WARN 4/15 network=bridge: outbound egress is possible; prefer network=none
Read-only root filesystem ❌ FAIL 0/10 root filesystem is writable: tamper/persistence surface
No docker.sock exposure ✅ PASS 15/15 no docker.sock / OCI control socket mounted
No shared host namespaces ✅ PASS 10/10 no host PID/IPC/network namespace sharing

Harden it: the highest-value fixes

Applying these to your docker run hazelcast targets the biggest gaps first (most points at stake first):

  • Dropped capabilities, --cap-drop=ALL
    Drop every Linux capability; add back only what the workload provably needs.
  • Network isolation / egress, --network=none
    Cut egress so a compromised workload cannot reach the network or exfiltrate.
  • Read-only root filesystem, --read-only --tmpfs /tmp
    Make the root filesystem read-only to remove the tamper/persistence surface.

The fixes above, plus the rest of IronClaw's recommended flag set, as one command:

docker run -d --name hazelcast-hardened \
  --user 65532:65532 \
  --cap-drop=ALL \
  --security-opt=no-new-privileges \
  --read-only --tmpfs /tmp \
  --network=none \
  hazelcast/hazelcast:5.5.0

This image has not been re-scanned under those flags, so this page states no hardened score for it. The one hardened run this survey measures is nginx:1.27-alpine, which reaches 100/100 (grade A) on exactly this flag set. Expect to adjust, and expect a ceiling: --network=none is not an option for a service that has to accept connections, and dropping it leaves the network-isolation dimension exactly where the table above has it. A container that writes outside /tmp will not boot read-only until you add a --tmpfs for each path it needs (some want a writable mode, e.g. --tmpfs /data:rw,mode=1777). Re-run ironctl scan on the result to see where yours actually lands.

Scan your own container

These grades come from ironctl scan, a single, credential-free command that audits any running container, docker-compose service, or Kubernetes manifest, not just this image:

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

# grade your own hazelcast the same way this page was generated
ironctl scan my-hazelcast

Badge this image

Maintain hazelcast (or run it)? Show its default-config isolation score with a badge that links back to this scorecard:

Container Isolation Score: 63/100 C

[![Container Isolation Score: 63/100 C](https://img.shields.io/badge/container%20isolation-63%2F100%20C-d4a72c)](https://ironsecco.github.io/ironclaw/scores/hazelcast/)

The badge is a plain shields.io URL: no server, no build step, nothing to host. It reflects this page's default-configuration grade. Hardened your own deployment? Generate a live badge of your config with ironctl scan --badge-json, or compare every image on the leaderboard.

Part of the Container Isolation Scores directory, default-configuration containment grades for the most-pulled public images.