Minimal prod delivery, authored directly on main (= prod's live revision d1a6628)
rather than by merging work/prod-standup-campaign. That branch carries the DNS apex
inversion, which would have set prod's BASE_DOMAIN to home.keithjackson.dev; prod
runs external-dns with policy: sync and txtOwnerId: prod, so it would have begun
rewriting apex records dev owns while the non-k3s prod still serves 443 there.
Operator sequencing is "dev migrates first". Nothing under clusters/ is touched
here at all, which is what makes this safe to land now.
Also deliberately EXCLUDED, because prod's live cluster-vars has no CNPG_BACKUP_PATH
key and StrictPostBuildSubstitutions would fail the whole tier: the R2 rework of
platform/velero/ and the CNPG destinationPath changes. Those land as
feat/r2-backup-prod together with the endpoint and bucket knobs.
WHAT THIS FIXES
1. Nothing was watching CNPG. All three prod databases have run since bootstrap with
ContinuousArchiving=False and firstRecoverabilityPoint=null, and no alert existed
anywhere. Because archive_mode=on and the archive command fails, Postgres refuses
to recycle WAL: keycloak-db holds every segment ever written (257 MiB on a 5 GiB
volume, ~6.8 MiB/h) — roughly 29 days to a disk-full PANIC on the cluster's SSO
database. Found by hand, 38h late. CNPGWALArchiveBacklog would have fired within
the hour, and evaluates to 3 firing series on prod today.
Rules key on cnpg_collector_* rather than cnpg_pg_stat_archiver_*, because
nextcloud-db's exporter cannot log in to its own database (permission denied for
database "nextcloud", both clusters) so the pg_stat_archiver family does not exist
for it — the obvious metric would have been blind to one database in three.
2. No PVC outside namespace=monitoring had any fill alert. New rules are partitioned
by namespace so they cannot double-fire with MonitoringVolumeFillingUp, pinned to
job="kubelet" (the apiserver copy of the same series reports namespace=default),
and guarded by inodes > 0 — block-mode PVCs such as unifi-rootdisk report
available=0 forever and would otherwise have fired critical permanently.
3. VPNTunnelDown could never clear on a cluster with no gateway node; it has been
firing critical on prod since 2026-09-05. Now guarded so prod self-arms the moment
a node carries the egress-gateway label, with dev's behaviour unchanged today.
4. The Keycloak realm import fails OPEN: an unresolved placeholder is stored
LITERALLY and the import still reports success, then sits at Done=True and never
re-runs. On 2026-09-06 all 12 prod OIDC clients held the literal
$(env:..._CLIENT_SECRET) string and every SSO login failed, while every signal was
green. The CronJob is the level-triggered correction; two alerts make its silence
observable. Repaired live already — this stops it recurring.
Also: the R-NET-04 conformance scan descended into nested agent worktrees under
.claude/, counting 4 egress policies where 1 is authored.
Co-Authored-By: Claude Opus 5 <[email protected]>
173 lines
7.7 KiB
YAML
173 lines
7.7 KiB
YAML
# =============================================================================
|
|
# L2 BEHAVIORAL — Default-deny east-west (R-NET-01) [RUNNABLE]
|
|
# Catalog: docs/14-testing-and-conformance.md "Default-deny (R-NET-01)"
|
|
# Spec: docs/03-networking-isolation-and-egress.md "Default-deny baseline"
|
|
#
|
|
# EXEMPLAR: this file is the reference for the tagging convention and step
|
|
# style. Other scaffolds follow the same shape (see tests/README.md).
|
|
#
|
|
# GIVEN a fresh workload in a namespace with the default-deny baseline applied,
|
|
# WHEN it attempts any in-cluster connection except DNS,
|
|
# THEN the connection FAILS (fail-closed); DNS still succeeds.
|
|
#
|
|
# This asserts observable BEHAVIOR (a probe Job cannot connect) via ephemeral
|
|
# probe pods, not the mere existence of a policy object (R-TEST-01).
|
|
#
|
|
# RUNNABLE ON THE BASE CLUSTER: the platform-wide default-deny baseline (T-2.2)
|
|
# is not yet applied cluster-wide, so this test applies an equivalent baseline
|
|
# as its OWN hermetic fixture in the per-test namespace. The fixture is a
|
|
# standard Kubernetes NetworkPolicy (Cilium enforces it) — portable and
|
|
# version-agnostic; the platform's baseline will be the CiliumNetworkPolicy
|
|
# equivalent, but the asserted BEHAVIOR (DNS-only egress) is identical, which
|
|
# is what R-TEST-01 requires us to prove. Chainsaw auto-creates/cleans the
|
|
# namespace (hermetic); the probe Jobs are ephemeral and self-cleaned with it.
|
|
# =============================================================================
|
|
apiVersion: chainsaw.kyverno.io/v1alpha1
|
|
kind: Test
|
|
metadata:
|
|
name: default-deny
|
|
labels:
|
|
# --- requirement tags (authoritative; one label per verified R-* ID) ---
|
|
conformance.home.k3s/R-NET-01: "true"
|
|
annotations:
|
|
conformance.home.k3s/requirements: "R-NET-01"
|
|
conformance.home.k3s/layer: "L2"
|
|
conformance.home.k3s/safety-critical: "false"
|
|
# executable => this test runs real behavior against a cluster (not a scaffold).
|
|
conformance.home.k3s/executable: "true"
|
|
conformance.home.k3s/intent: >-
|
|
Given a bare workload in a namespace with the default-deny baseline, only
|
|
DNS egress succeeds; every other in-cluster connection is denied.
|
|
spec:
|
|
steps:
|
|
# -- GIVEN: apply the default-deny baseline to the per-test namespace. Deny all
|
|
# ingress + egress for every pod; allow ONLY DNS egress to the in-cluster
|
|
# resolver. This is the hermetic stand-in for the platform baseline (T-2.2).
|
|
- name: given-default-deny-baseline
|
|
try:
|
|
- apply:
|
|
resource:
|
|
apiVersion: networking.k8s.io/v1
|
|
kind: NetworkPolicy
|
|
metadata:
|
|
name: default-deny-baseline
|
|
spec:
|
|
podSelector: {}
|
|
policyTypes: [Ingress, Egress]
|
|
# no ingress rules => deny all ingress
|
|
ingress: []
|
|
# only DNS egress to kube-dns (CoreDNS) in kube-system
|
|
egress:
|
|
- to:
|
|
- namespaceSelector:
|
|
matchLabels:
|
|
kubernetes.io/metadata.name: kube-system
|
|
podSelector:
|
|
matchLabels:
|
|
k8s-app: kube-dns
|
|
ports:
|
|
- protocol: UDP
|
|
port: 53
|
|
- protocol: TCP
|
|
port: 53
|
|
|
|
# -- THEN(DNS allowed): DNS egress MUST succeed even under default-deny. An
|
|
# ephemeral Job resolves an in-cluster name; we assert the Job SUCCEEDS.
|
|
# (Chainsaw polls the Job status within the assert timeout — bounded, no sleeps.)
|
|
- name: then-dns-succeeds
|
|
try:
|
|
- apply:
|
|
resource:
|
|
apiVersion: batch/v1
|
|
kind: Job
|
|
metadata:
|
|
name: dns-probe
|
|
spec:
|
|
backoffLimit: 4
|
|
activeDeadlineSeconds: 120
|
|
template:
|
|
metadata:
|
|
labels: { app: conformance-probe }
|
|
spec:
|
|
restartPolicy: Never
|
|
containers:
|
|
- name: probe
|
|
image: busybox:1.36.1
|
|
command:
|
|
- sh
|
|
- -c
|
|
- |
|
|
set -eu
|
|
# DNS is the one egress the baseline permits.
|
|
nslookup kubernetes.default.svc.cluster.local
|
|
echo DNS_OK
|
|
- assert:
|
|
resource:
|
|
apiVersion: batch/v1
|
|
kind: Job
|
|
metadata: { name: dns-probe }
|
|
status: { succeeded: 1 }
|
|
|
|
# -- THEN(everything else denied): a non-DNS in-cluster connection MUST fail
|
|
# closed. The Job attempts to reach the kube-apiserver ClusterIP (always
|
|
# present) on 443; the probe EXITS 0 only when that connection is BLOCKED,
|
|
# so a green Job proves the deny. A leak (connection succeeds) fails the Job.
|
|
- name: then-nondns-denied
|
|
try:
|
|
- apply:
|
|
resource:
|
|
apiVersion: batch/v1
|
|
kind: Job
|
|
metadata:
|
|
name: block-probe
|
|
spec:
|
|
backoffLimit: 4
|
|
activeDeadlineSeconds: 120
|
|
template:
|
|
metadata:
|
|
labels: { app: conformance-probe }
|
|
spec:
|
|
restartPolicy: Never
|
|
containers:
|
|
- name: probe
|
|
image: busybox:1.36.1
|
|
command:
|
|
- sh
|
|
- -c
|
|
- |
|
|
set -u
|
|
# DNS resolves (allowed); the TCP connect must be dropped.
|
|
if nc -w 4 kubernetes.default.svc.cluster.local 443 </dev/null 2>/dev/null; then
|
|
echo "LEAK: reached kube-apiserver under default-deny" 1>&2
|
|
exit 1
|
|
fi
|
|
echo DENIED_OK
|
|
- assert:
|
|
resource:
|
|
apiVersion: batch/v1
|
|
kind: Job
|
|
metadata: { name: block-probe }
|
|
status: { succeeded: 1 }
|
|
# -- chainsaw deletes the namespace (policy + probe Jobs) automatically; this
|
|
# block is the explicit hook for any out-of-namespace fixtures (none here).
|
|
finally:
|
|
- script:
|
|
content: 'echo "cleanup handled by chainsaw namespace teardown"'
|
|
# -----------------------------------------------------------------------------------
|
|
# PROOF-OF-PASS SENTINEL (R-SEC-03). Every step above is an assert/Job-status gate, so
|
|
# chainsaw already fails the test before reaching this step if any of them fails — but a
|
|
# green Chainsaw JUnit carries NO stdout, and the L2/L3 tests in this suite can self-skip
|
|
# GREEN, so lib/record_run.py refuses to treat a bare `passed` as evidence and records
|
|
# `unknown-no-sentinel-declared` instead. Without a declared token this test could pass
|
|
# forever and NEVER become a recorded proof of R-NET-01 — the isolation half of docs/07's audit proof list
|
|
# — which is precisely what docs/07 asks for ("These proofs are part of the platform
|
|
# acceptance suite, not one-time checks"). This step exists only to emit that token, and it
|
|
# is reached only when everything above has passed. Asserted by
|
|
# tests/l0-static/auditability-proofs-producible.
|
|
- name: then-emit-the-proof-of-pass-sentinel
|
|
try:
|
|
- script:
|
|
content: echo DEFAULT_DENY_PROVEN_OK
|
|
check:
|
|
(contains($stdout, 'DEFAULT_DENY_PROVEN_OK')): true
|