How to harden a ScyllaDB container (and is scylla:6.2 safe for untrusted workloads?)¶
Short answer: a stock docker run scylladb/scylla:6.2 is not a boundary you should trust
around untrusted code or an untrusted network. Graded on IronClaw's seven-dimension containment
scale, the default configuration scores 48 of 100, grade D (porous). Higher is safer. Four
runtime flags take the same image to 100 of 100, grade A. This guide shows the exact gaps and
the exact fixes, straight from the scan data.
Every number here comes from a read-only
docker inspectofscylladb/scylla:6.2, 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 scylladb/scylla:6.2, four of them fail or warn:
| Dimension | Verdict | Score | What the scan found |
|---|---|---|---|
| Non-root user (uid != 0) | ❌ FAIL | 0/15 | runs as root (uid 0); a container escape starts with host-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 |
For a wide-column database, the two that should worry you most are egress and root. A Scylla process that can reach the network is a Scylla process that can exfiltrate every row it stores the moment it is compromised (a poisoned UDF, a CQL or driver CVE). And a root process that escapes the container escapes as root on the host.
Harden it: the exact --fix remediation¶
ironctl scan my-scylla --fix prints one remediation per failed dimension, then a single
copy-pasteable hardened run. For scylladb/scylla:6.2:
--cap-drop=ALL(Dropped capabilities, +16): drop every Linux capability; add back only what the workload provably needs. Scylla itself needs none of the defaults.--user 65532:65532(Non-root user, +15): pin a non-root uid so an escape does not begin as host uid 0. Point the data directory at a volume this uid owns.--network=none(Network isolation, +11): for a single-node store reached only by co-located services on a private user-defined network, cut host egress entirely; otherwise attach a single internal network with no default route, notbridge. (A multi-node cluster needs inter-node gossip on a scoped private network, which is a WARN, not a clean pass, see the ceiling note below.)--read-only --tmpfs /tmp(Read-only rootfs, +10): make the root filesystem read-only and mount/var/lib/scyllaas an explicit writable volume. Removes the persistence surface.
Before and after¶
# Before: 48/100, grade D
docker run -d --name scylla scylladb/scylla:6.2
# After: 100/100, grade A (single-node)
docker run -d --name scylla-hardened \
--user 65532:65532 \
--cap-drop=ALL \
--security-opt=no-new-privileges \
--read-only --tmpfs /tmp \
-v scylla-data:/var/lib/scylla \
--network=none \
scylladb/scylla:6.2
Rescan and the same seven dimensions all pass: ironctl scan scylla-hardened reports
100/100 grade A. That is a 52-point swing from four one-line flags, no image rebuild.
The honest ceiling for a multi-node cluster¶
The 100/100 above is a single-node store with network=none. A multi-node Scylla cluster
cannot use network=none: nodes gossip and stream data to each other. Attach a private
user-defined network scoped to just the cluster with no default route out, and the network
dimension scores 4 of 15 (a WARN, not a fail) instead of the full 15. That puts a hardened
multi-node cluster at 89 of 100, grade B, one point off an A. The other six dimensions still
max out. That is the honest ceiling when nodes must talk, and it is a long way from the default D.
Verify it on your own database¶
The grade above is the default image. Your deployment is what matters. Scan it in ten seconds:
# install (Homebrew)
brew install ironsecco/ironclaw/ironclaw
# grade your running container, then print the fixes
ironctl scan my-scylla
ironctl scan my-scylla --fix
ironctl scan also reads a docker-compose.yml service or a Kubernetes manifest, so you can
grade the Scylla in your stack, not just a bare docker run.
Keep going¶
- All hardening guides →: every harden-a-container walkthrough, with grade deltas.
- scylla:6.2 isolation scorecard →: the full dimension breakdown.
- Databases, ranked by isolation →: how Scylla compares to Cassandra, MongoDB, Postgres, and the rest.
- How to harden a Cassandra container →: the same walkthrough for the wide-column database Scylla is compatible with.
- Scan any container in 10 seconds →: the full
ironctl scanreference. - Run untrusted code in a real sandbox →: IronClaw wraps every AI-agent session in a gVisor/Kata boundary with
network=noneby default.