chore: add integration tests for image-cache

Provide separate `integration/image-cache` tag.

Closes siderolabs#9860

Signed-off-by: Dmitriy Matrenichev <[email protected]>

.github/workflows/ci.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-11-28T14:18:12Z by kres 232fe63.
+# Generated on 2024-12-03T22:25:05Z by kres 232fe63.
 
 name: default
 concurrency:
@@ -1609,6 +1609,99 @@ jobs:
           TF_SCRIPT_DIR: _out/contrib
         run: |
           make e2e-cloud-tf
+  integration-image-cache:
+    permissions:
+      actions: read
+      contents: write
+      issues: read
+      packages: write
+      pull-requests: read
+    runs-on:
+      - self-hosted
+      - talos
+    if: contains(fromJSON(needs.default.outputs.labels), 'integration/image-cache')
+    needs:
+      - default
+    steps:
+      - name: gather-system-info
+        id: system-info
+        uses: kenchan0130/[email protected]
+        continue-on-error: true
+      - name: print-system-info
+        run: |
+          MEMORY_GB=$((${{ steps.system-info.outputs.totalmem }}/1024/1024/1024))
+
+          OUTPUTS=(
+            "CPU Core: ${{ steps.system-info.outputs.cpu-core }}"
+            "CPU Model: ${{ steps.system-info.outputs.cpu-model }}"
+            "Hostname: ${{ steps.system-info.outputs.hostname }}"
+            "NodeName: ${NODE_NAME}"
+            "Kernel release: ${{ steps.system-info.outputs.kernel-release }}"
+            "Kernel version: ${{ steps.system-info.outputs.kernel-version }}"
+            "Name: ${{ steps.system-info.outputs.name }}"
+            "Platform: ${{ steps.system-info.outputs.platform }}"
+            "Release: ${{ steps.system-info.outputs.release }}"
+            "Total memory: ${MEMORY_GB} GB"
+          )
+
+          for OUTPUT in "${OUTPUTS[@]}";do
+            echo "${OUTPUT}"
+          done
+        continue-on-error: true
+      - name: checkout
+        uses: actions/checkout@v4
+      - name: Unshallow
+        run: |
+          git fetch --prune --unshallow
+      - name: Set up Docker Buildx
+        id: setup-buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          driver: remote
+          endpoint: tcp://buildkit-amd64.ci.svc.cluster.local:1234
+        timeout-minutes: 10
+      - name: Download artifacts
+        if: github.event_name != 'schedule'
+        uses: actions/download-artifact@v4
+        with:
+          name: talos-artifacts
+          path: _out
+      - name: Fix artifact permissions
+        if: github.event_name != 'schedule'
+        run: |
+          xargs -a _out/executable-artifacts -I {} chmod +x {}
+      - name: ci-temp-release-tag
+        if: github.event_name != 'schedule'
+        run: |
+          make ci-temp-release-tag
+      - name: uki-certs
+        if: github.event_name == 'schedule'
+        env:
+          PLATFORM: linux/amd64
+        run: |
+          make uki-certs
+      - name: image-cache
+        env:
+          IMAGE_REGISTRY: registry.dev.siderolabs.io
+          PLATFORM: linux/amd64,linux/arm64
+          PUSH: "true"
+        run: |
+          make cache-create
+      - name: use-image-cache
+        env:
+          GITHUB_STEP_NAME: ${{ github.job}}-e2e-image-cache
+          IMAGE_REGISTRY: registry.dev.siderolabs.io
+        run: |
+          sudo -E make e2e-image-cache
+      - name: save artifacts
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: talos-logs-integration-image-cache
+          path: |-
+            /tmp/logs-*.tar.gz
+            /tmp/support-*.zip
+          retention-days: "5"
   integration-image-factory:
     permissions:
       actions: read

.kres.yaml

@@ -72,6 +72,7 @@ spec:
     - integration-images
     - integration-reproducibility-test
     - integration-cloud-images
+    - integration-image-cache
     - integration-image-factory
     - integration-aws
     - integration-aws-nvidia-oss
@@ -1509,6 +1510,54 @@ spec:
             PLATFORM: linux/amd64,linux/arm64
             IMAGE_REGISTRY: registry.dev.siderolabs.io
         - name: cloud-images
+    - name: integration-image-cache
+      buildxOptions:
+        enabled: true
+      depends:
+        - default
+      runners:
+        - self-hosted
+        - talos
+      triggerLabels:
+        - integration/image-cache
+      steps:
+        - name: download-artifacts
+          conditions:
+            - not-on-schedule
+          artifactStep:
+            type: download
+            artifactName: talos-artifacts
+            artifactPath: _out
+        - name: ci-temp-release-tag
+          conditions:
+            - not-on-schedule
+        - name: uki-certs
+          conditions:
+            - only-on-schedule
+          environment:
+            PLATFORM: linux/amd64
+        - name: image-cache
+          command: cache-create
+          environment:
+            PLATFORM: linux/amd64,linux/arm64
+            IMAGE_REGISTRY: registry.dev.siderolabs.io
+            PUSH: true
+        - name: use-image-cache
+          command: e2e-image-cache
+          withSudo: true
+          environment:
+            IMAGE_REGISTRY: registry.dev.siderolabs.io
+            GITHUB_STEP_NAME: ${{ github.job}}-e2e-image-cache
+        - name: save-talos-logs
+          conditions:
+            - always
+          artifactStep:
+            type: upload
+            artifactName: talos-logs-integration-image-cache
+            disableExecutableListGeneration: true
+            artifactPath: /tmp/logs-*.tar.gz
+            additionalArtifacts:
+              - "/tmp/support-*.zip"
     - name: integration-image-factory
       buildxOptions:
         enabled: true

Makefile

@@ -459,6 +459,12 @@ uki-certs: talosctl ## Generate test certificates for SecureBoot/PCR Signing
 	@$(TALOSCTL_EXECUTABLE) gen secureboot pcr
 	@$(TALOSCTL_EXECUTABLE) gen secureboot database
 
+.PHONY: cache-create
+cache-create: installer imager ## Generate image cache.
+	@$(TALOSCTL_EXECUTABLE) images default | grep -v 'siderolabs | $(TALOSCTL_EXECUTABLE) images cache-create --image-cache-path=/tmp/cache.tar --images=- --force
+	@crane push /tmp/cache.tar $(REGISTRY_AND_USERNAME)/image-cache
+	@$(MAKE) image-iso IMAGER_ARGS="--image-cache=$(REGISTRY_AND_USERNAME)/image-cache:latest --extra-kernel-arg='console=ttyS0'"
+
 # Code Quality
 
 api-descriptors: ## Generates API descriptors used to detect breaking API changes.

hack/test/e2e-image-cache.sh

@@ -0,0 +1,58 @@
+#!/usr/bin/env bash
+
+set -eoux pipefail
+
+source ./hack/test/e2e.sh
+
+PROVISIONER=qemu
+CLUSTER_NAME=e2e-image-cache
+
+NODE="172.20.2.2"
+
+function build_special_mirrors {
+  REGISTRY_MIRROR_FLAGS=()
+
+  for registry in docker.io quay.io ghcr.io; do
+    local service="registry-${registry//./-}.ci.svc"
+    addr=$(python3 -c "import socket; print(socket.gethostbyname('${service}'))")
+
+    REGISTRY_MIRROR_FLAGS+=("--registry-mirror=${registry}=http://${addr}:5000")
+  done
+}
+
+function create_cluster {
+  build_special_mirrors
+
+  "${TALOSCTL}" cluster create \
+    --provisioner="${PROVISIONER}" \
+    --name="${CLUSTER_NAME}" \
+    --kubernetes-version=${KUBERNETES_VERSION} \
+    --iso-path=${ARTIFACTS}/metal-amd64.iso \
+    --controlplanes=1 \
+    --workers=0 \
+    --mtu=1430 \
+    --memory=2048 \
+    --cpus=2.0 \
+    --cidr=172.20.2.0/24 \
+    --with-apply-config \
+    --config-patch=@hack/test/patches/image-cache.yaml \
+    --install-image=${REGISTRY:-ghcr.io}/siderolabs/installer:${TAG} \
+    --cni-bundle-url=${ARTIFACTS}/talosctl-cni-bundle-'${ARCH}'.tar.gz \
+    "${REGISTRY_MIRROR_FLAGS[@]}" \
+    "${QEMU_FLAGS[@]}"
+
+  "${TALOSCTL}" config node "${NODE}"
+}
+
+function destroy_cluster() {
+  "${TALOSCTL}" cluster destroy \
+    --name "${CLUSTER_NAME}" \
+    --provisioner "${PROVISIONER}" \
+    --save-cluster-logs-archive-path="/tmp/logs-${CLUSTER_NAME}.tar.gz" \
+    --save-support-archive-path="/tmp/support-${CLUSTER_NAME}.zip"
+}
+
+trap destroy_cluster SIGINT EXIT
+
+create_cluster
+sleep 5

hack/test/patches/image-cache.yaml

@@ -0,0 +1,25 @@
+machine:
+  features:
+    imageCache:
+      localEnabled: true
+  registries:
+    mirrors:
+      gcr.io:
+        skipFallback: true
+        endpoints:
+        - http://172.20.0.250:65000
+      k8s.gcr.io:
+        skipFallback: true
+        endpoints:
+        - http://172.20.0.250:65000
+      registry.k8s.io:
+        skipFallback: true
+        endpoints:
+        - http://172.20.0.250:65000
+---
+apiVersion: v1alpha1
+kind: VolumeConfig
+name: IMAGECACHE
+provisioning:
+  diskSelector:
+    match: 'system_disk'