ELares · Jul 24, 2026 · Jul 24, 2026
@@ -27,14 +27,14 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Create a kind cluster
-        uses: helm/[email protected]
+        uses: helm/kind-action@0025e74a8c7512023d06dc019c617aa3cf561fde # v1.10.0
         with:
           # Pin the k8s version to 1.30 so the run is deterministic AND exercises the native
           # SleepAction preStop path (beta/default-on from 1.30; the chart gates it on >=1.30).
           # The action's default is 1.29, where SleepAction is alpha-off -- a moving, wrong target.
           version: v0.23.0
           node_image: kindest/node:v1.30.0
-      - uses: azure/setup-helm@v4
+      - uses: azure/setup-helm@1a275c3b69536ee54be43f2070a358922e12c8d4 # v4.3.1
         with:
           version: ${{ env.HELM_VERSION }}
       - name: Install (standalone) + wait + helm test

@@ -38,11 +38,11 @@ jobs:
       id-token: write # keyless cosign signing via OIDC
     steps:
       - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-      - uses: azure/setup-helm@v4
+      - uses: azure/setup-helm@1a275c3b69536ee54be43f2070a358922e12c8d4 # v4.3.1
         with:
           version: "v3.15.4"
       - name: Install cosign
-        uses: sigstore/[email protected]
+        uses: sigstore/cosign-installer@dc72c7d5c4d10cd6bcb8cf6e3fd625a9e5e537da # v3.7.0
       - name: helm lint + package
         id: pkg
         run: |

@@ -40,7 +40,7 @@ jobs:
     timeout-minutes: 10
     steps:
       - uses: actions/checkout@v4
-      - uses: azure/setup-helm@v4
+      - uses: azure/setup-helm@1a275c3b69536ee54be43f2070a358922e12c8d4 # v4.3.1
         with:
           version: ${{ env.HELM_VERSION }}
       - name: helm lint (defaults + console enabled)

@@ -60,9 +60,11 @@ not stale text). The chart is production-safe for **fixed-size** clusters.
   on its first run it caught + fixed a real shipped preStop-gate bug (SleepAction is
   beta/default-on only from **1.30**, not 1.29), but `ct` and an **upgrade**-path test are still
   missing. The chart is now 0.3.0.
-- **Genuinely remaining** (each also flagged in the Section 2 table):
-  1. **P2-8** the doc-only clarifications (incl. pinning the busybox init image by digest).
-  2. **The day-2 scaling Operator** (Section 5) -- a deliberate non-goal for now; the chart + the
+- **P2 is now COMPLETE (P2-1..P2-8)**: P2-3 finished with the upgrade-path e2e (#763), P2-6 the
+  ServiceMonitor scrape knobs + the alert rules as a PrometheusRule (#764), P2-7 the console-UI
+  Ingress (#765), P2-8 the doc clarifications + the opt-in busybox digest pin.
+- **Genuinely remaining** (also flagged in the Section 2 table):
+  1. **The day-2 scaling Operator** (Section 5) -- a deliberate non-goal for now; the chart + the
      #392 upgrade driver cover fixed-size clusters.
 
 ---

@@ -29,7 +29,7 @@ name: ironcache-cluster
 
 services:
   ironcache-1:
-    image: ghcr.io/elares/ironcache:latest
+    image: ghcr.io/elares/ironcache:${IRONCACHE_VERSION:-0.1.0}
     hostname: ironcache-1
     command:
       - server
@@ -60,7 +60,7 @@ services:
     restart: unless-stopped
 
   ironcache-2:
-    image: ghcr.io/elares/ironcache:latest
+    image: ghcr.io/elares/ironcache:${IRONCACHE_VERSION:-0.1.0}
     hostname: ironcache-2
     command:
       - server
@@ -85,7 +85,7 @@ services:
     restart: unless-stopped
 
   ironcache-3:
-    image: ghcr.io/elares/ironcache:latest
+    image: ghcr.io/elares/ironcache:${IRONCACHE_VERSION:-0.1.0}
     hostname: ironcache-3
     command:
       - server

@@ -23,7 +23,7 @@ name: ironcache-console
 
 # The common console replica config (both replicas are identical).
 x-console: &console
-  image: ghcr.io/elares/ironcache-console:latest
+  image: ghcr.io/elares/ironcache-console:${IRONCACHE_VERSION:-0.1.0}
   command: [run]
   environment:
     IRONCACHE_CONSOLE_HTTP_ADDR: 0.0.0.0:9180

@@ -18,7 +18,7 @@ name: ironcache
 
 services:
   ironcache:
-    image: ghcr.io/elares/ironcache:latest
+    image: ghcr.io/elares/ironcache:${IRONCACHE_VERSION:-0.1.0}
     command:
       - server
       - --bind

@@ -103,7 +103,8 @@ spec:
         # ordinal) into /run/ironcache/ironcache.toml, which the main container
         # then passes to --config. Keeps the runtime image distroless/shell-free.
         - name: stamp-identity
-          image: "{{ .Values.initImage.repository }}:{{ .Values.initImage.tag }}"
+          {{- /* digest wins when set (supply-chain pin); note the `@` form, not `:`. */}}
+          image: "{{ .Values.initImage.repository }}{{ if .Values.initImage.digest }}@{{ .Values.initImage.digest }}{{ else }}:{{ .Values.initImage.tag }}{{ end }}"
           imagePullPolicy: {{ .Values.initImage.pullPolicy }}
           command: ["/bin/sh", "/scripts/stamp-identity.sh"]
           securityContext:

@@ -567,6 +567,10 @@
             "Never"
           ],
           "description": "Init image pull policy."
+        },
+        "digest": {
+          "type": "string",
+          "description": "Optional image digest (sha256:...) for the init image; when set it WINS over tag and renders repo@digest. Empty by default so air-gapped pre-loaded images still work."
         }
       }
     },

@@ -57,7 +57,12 @@ persistence:
   # The PVC mount path == data_dir.
   mountPath: /var/lib/ironcache
   size: 10Gi
-  # storageClassName: "" leaves the cluster default; set an SSD class in prod.
+  # storageClassName: "" leaves the cluster default; set an SSD class in prod. NOTE the PV
+  # RECLAIM POLICY comes from the StorageClass, not this chart: for production use a class whose
+  # reclaimPolicy is `Retain`, so deleting the PVC does not also destroy the underlying volume
+  # (the chart's persistentVolumeClaimRetentionPolicy protects the PVC; the StorageClass protects
+  # the PV behind it). Also: labels/annotations on a volumeClaimTemplate are applied at PVC
+  # CREATION only -- editing them later does NOT propagate to already-created PVCs.
   storageClassName: ""
   accessModes:
     - ReadWriteOnce
@@ -289,6 +294,9 @@ pdb:
 # compatible with `restricted` because it needs no in-image binary.
 #
 # The image already runs as nonroot (uid 65532); fsGroup makes the PVC writable by it.
+# runAsUser/runAsGroup MUST match the image's baked-in UID/GID (65532). Changing them to an id
+# the image does not own makes the process unable to read its own files (and, with fsGroup left
+# behind, the PVC too) -- change these only if you also rebuild the image.
 podSecurityContext:
   runAsNonRoot: true
   runAsUser: 65532
@@ -302,6 +310,9 @@ podSecurityContext:
     type: RuntimeDefault
 containerSecurityContext:
   allowPrivilegeEscalation: false
+  # readOnlyRootFilesystem applies to `/` ONLY. The pod still has the writable mounts it needs:
+  # the data PVC at persistence.mountPath and a small emptyDir at /run/ironcache that the init
+  # container writes the stamped config into. That combination is by design, not a leak.
   readOnlyRootFilesystem: true
   capabilities:
     drop:
@@ -378,6 +389,14 @@ topologySpread:
 initImage:
   repository: busybox
   tag: "1.36"
+  # SUPPLY-CHAIN HARDENING (opt-in): pin the init image by DIGEST instead of the mutable tag.
+  # When set it WINS over `tag` and renders `busybox@sha256:...`. The verified multi-arch
+  # (amd64 + arm64 + more) manifest-list digest for busybox:1.36 is:
+  #   sha256:73aaf090f3d85aa34ee199857f03fa3a95c8ede2ffd4cc2cdb5b94e566b11662
+  # Left EMPTY by default deliberately: an AIR-GAPPED cluster pre-loads its own busybox tarball
+  # (see deploy/K3S.md), whose digest will differ, and a digest pin would then fail the pull.
+  # Set it wherever you control the registry.
+  digest: ""
   pullPolicy: IfNotPresent
 
 # --- Console (epic #352, issue #363) ------------------------------------------