ci(internal-workflows): make release publishing recoverable

.github/workflows/publish.yaml

@@ -6,9 +6,15 @@ on:
   release:
     types:
     - published
+  workflow_dispatch:
+    inputs:
+      release_tag:
+        description: Existing release tag to publish.
+        required: true
+        type: string
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.event.release.tag_name }}
+  group: ${{ github.workflow }}-${{ github.event.release.tag_name || inputs.release_tag }}
   cancel-in-progress: true
 
 env:
@@ -17,8 +23,8 @@ env:
   image_path: images/homelab-workspace
   template_path: templates/kubernetes/homelab-workspace
   template_name: homelab-workspace
-  source_ref: ${{ github.event.release.tag_name }}
-  source_sha: ${{ github.event.release.tag_name }}
+  source_ref: ${{ github.event.release.tag_name || inputs.release_tag }}
+  source_sha: ${{ github.event.release.tag_name || inputs.release_tag }}
 
 jobs:
   publish-image:
@@ -33,7 +39,7 @@ jobs:
       platforms: linux/amd64,linux/arm64
       private_registry_repository: ${{ vars.CONTAINER_REGISTRY_PATH }}/coder-workspace
       private_registry_build_cache: ${{ vars.CONTAINER_REGISTRY_CACHE_PATH }}/coder-workspace
-      git_ref: ${{ github.event.release.tag_name }}
+      git_ref: ${{ github.event.release.tag_name || inputs.release_tag }}
       ghcr_repository: ppat/coder-workspace
       timeout_minutes: 180
     secrets:
@@ -50,11 +56,20 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 15
     steps:
-    - name: Checkout
+    - name: Checkout publishing automation
       uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       with:
         fetch-depth: 1
         persist-credentials: false
+        path: .automation
+        ref: main
+
+    - name: Checkout release source
+      uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+      with:
+        fetch-depth: 1
+        persist-credentials: false
+        path: release
         ref: ${{ env.source_ref }}
 
     - name: Tailscale Connect
@@ -67,16 +82,16 @@ jobs:
         version: "1.102.4"
 
     - name: Login to Coder
-      uses: ./actions/coder-cli-login
+      uses: ./.automation/actions/coder-cli-login
       with:
         coder_url: ${{ secrets.CODER_URL }}
         coder_email: ${{ secrets.CODER_EMAIL }}
         coder_password: ${{ secrets.CODER_PASSWORD }}
 
     - name: Publish template
-      uses: ./actions/coder-template-push
+      uses: ./.automation/actions/coder-template-push
       with:
-        template_dir: ${{ env.template_path }}
+        template_dir: release/${{ env.template_path }}
         template_name: ${{ env.template_name }}
         template_version: ${{ env.source_sha }}
         workspace_image: "ghcr.io/ppat/coder-workspace:${{ needs.publish-image.outputs.image_tag }}"

actions/coder-template-push/action.yaml

@@ -54,6 +54,31 @@ runs:
       TEMPLATE_MESSAGE: ${{ inputs.message }}
     # yamllint disable-line rule:indentation
     run: |
+      existing_version=""
+      if coder templates list --output json | jq --exit-status --arg name "${TEMPLATE_NAME}" \
+        'any(.[]; .Template.name == $name)' >/dev/null; then
+        existing_version="$(coder templates versions list "${TEMPLATE_NAME}" --include-archived --output json | jq --compact-output \
+          --arg name "${TEMPLATE_VERSION}" '.[] | select(.TemplateVersion.name == $name)')"
+      fi
+      if [[ -n "${existing_version}" ]]; then
+        existing_status="$(jq --raw-output '.TemplateVersion.job.status' <<<"${existing_version}")"
+        existing_active="$(jq --raw-output '.active' <<<"${existing_version}")"
+        existing_id="$(jq --raw-output '.TemplateVersion.id' <<<"${existing_version}")"
+        if [[ "${existing_status}" != "failed" || "${existing_active}" != "false" ]]; then
+          echo "::error::Template version ${TEMPLATE_VERSION} already exists with status=${existing_status}, active=${existing_active}; refusing to replace it."
+          exit 1
+        fi
+
+        coder_url="$(<"${HOME}/.config/coderv2/url")"
+        coder_session_token="$(<"${HOME}/.config/coderv2/session")"
+        failed_version_name="${TEMPLATE_VERSION}-failed-${existing_id%%-*}"
+        curl --fail --silent --show-error --request PATCH \
+          --header "Coder-Session-Token: ${coder_session_token}" \
+          --header "Content-Type: application/json" \
+          --data "$(jq --null-input --arg name "${failed_version_name}" '{name: $name}')" \
+          "${coder_url}/api/v2/templateversions/${existing_id}" >/dev/null
+      fi
+
       variables_file="${RUNNER_TEMP}/coder-template-variables.yaml"
       jq --null-input \
         --arg kubernetes_config_path "${KUBERNETES_CONFIG_PATH}" \