Skip to content

How to harden a Qdrant container (and is qdrant:v1.12.4 safe for untrusted workloads?)

Short answer: a stock docker run qdrant/qdrant:v1.12.4 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 inspect of qdrant/qdrant:v1.12.4, 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 qdrant/qdrant:v1.12.4, 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 vector database, the two that should worry you most are egress and root. A Qdrant process that can reach the network is a Qdrant process that can exfiltrate every embedding and its payload the moment it is compromised (a poisoned collection, a REST or gRPC CVE). And a root process that escapes the container escapes as root on the host.

Harden it: the exact --fix remediation

ironctl scan my-qdrant --fix prints one remediation per failed dimension, then a single copy-pasteable hardened run. For qdrant/qdrant:v1.12.4:

  • --cap-drop=ALL (Dropped capabilities, +16): drop every Linux capability; add back only what the workload provably needs. Qdrant 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 storage directory at a volume this uid owns.
  • --network=none (Network isolation, +11): if the vector store is only reached by services on a private user-defined network, cut host egress entirely; otherwise attach a single internal network with no default route, not bridge.
  • --read-only --tmpfs /tmp (Read-only rootfs, +10): make the root filesystem read-only and mount /qdrant/storage as an explicit writable volume. Removes the persistence surface.

Before and after

# Before: 48/100, grade D
docker run -d --name qdrant qdrant/qdrant:v1.12.4

# After: 100/100, grade A
docker run -d --name qdrant-hardened \
  --user 65532:65532 \
  --cap-drop=ALL \
  --security-opt=no-new-privileges \
  --read-only --tmpfs /tmp \
  -v qdrant-data:/qdrant/storage \
  --network=none \
  qdrant/qdrant:v1.12.4

Rescan and the same seven dimensions all pass: ironctl scan qdrant-hardened reports 100/100 grade A. That is a 52-point swing from four one-line flags, no image rebuild.

Verify it on your own vector store

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-qdrant
ironctl scan my-qdrant --fix

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

Keep going