ci: build every template's image independently, skip only failed ones

Fixes a regression where the whole provision job (all 6 templates) was
skipped whenever build-web-image failed, since Actions skips a job whose
needs: dependency failed by default. Generalizes the single web-image
build step into a loop over every templates/*/Dockerfile, building and
pushing each independently so one failure doesn't block the rest, and
reports failures via a job output. provision now runs unconditionally
(if: always()) and skips pushing only the specific template(s) whose
image build failed this run, leaving their previous working version in
place instead of pushing one with no matching registry tag.

.gitea/workflows/coder-templates.yml

@@ -14,18 +14,25 @@ name: Provision Coder Templates
 #   CODER_SESSION_TOKEN   a token from `coder tokens create`, ideally under a
 #                          dedicated service account rather than a personal one
 #   GITEA_PACKAGE_TOKEN   a Gitea access token (user Settings > Applications)
-#                          with write:package scope, for pushing templates/web's
+#                          with write:package scope, for pushing each
-#                          image to this instance's container registry. Only
+#                          Dockerfile-having template's image to this
-#                          the octoturge account's own token is used - login()
+#                          instance's container registry. Only the octoturge
+#                          account's own token is used - docker login below
 #                          hardcodes that username to match.
 #
-# templates/web builds its Docker image here (build-web-image, on the
+# Any templates/<env>/ that has its own Dockerfile gets its image built and
-# dedicated "docker-build" runner - see templates/web/main.tf for why: that
+# pushed here (build-images, on the dedicated "docker-build" runner - see
-# runner is scoped to this repo only and has host Docker socket access that
+# templates/web/main.tf for why: that runner is scoped to this repo only and
-# the shared runner-1 deliberately doesn't). provision then just pulls the
+# has host Docker socket access that the shared runner-1 deliberately
-# tag build-web-image produced, instead of building it itself at
+# doesn't). provision then just pulls the tag build-images produced, instead
-# `terraform apply` time - keeps the slow Rust toolchain compile off of
+# of building it itself at `terraform apply` time - keeps a slow toolchain
-# "someone is waiting to create a workspace".
+# compile off of "someone is waiting to create a workspace".
+#
+# Each template's image is built independently - one Dockerfile failing to
+# build doesn't stop the others from building, and provision skips pushing
+# only the specific template(s) whose image build failed this run (leaving
+# their previous, already-working Coder template version in place) rather
+# than skipping every template or pushing one with no matching image.
 
 on:
   push:
@@ -36,35 +43,62 @@ on:
   workflow_dispatch: {}
 
 jobs:
-  build-web-image:
+  build-images:
     # The docker-build runner's docker_host: "" setting already auto-injects
     # /var/run/docker.sock into job containers - an explicit
     # container.volumes mount for the same path here fails at container
     # creation with "Duplicate mount point: /var/run/docker.sock".
     runs-on: docker-build
+    outputs:
+      failed_templates: ${{ steps.build.outputs.failed_templates }}
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
-      - name: Build and push templates/web's image (skips if the tag already exists)
+      - name: Build and push every template's image (skips a tag that's already in the registry)
+        id: build
         run: |
           set -e
-          TAG="$(sha1sum templates/web/Dockerfile | cut -d' ' -f1)"
-          IMAGE="git.octoturge.com/octo-tech/profiles-web:${TAG}"
           echo "${{ secrets.GITEA_PACKAGE_TOKEN }}" | docker login git.octoturge.com -u octoturge --password-stdin
-          if docker manifest inspect "$IMAGE" >/dev/null 2>&1; then
+
-            echo "$IMAGE already in the registry (Dockerfile unchanged), skipping build."
+          FAILED=""
-            exit 0
+          for dockerfile in templates/*/Dockerfile; do
-          fi
+            [ -e "$dockerfile" ] || continue
-          docker build -t "$IMAGE" templates/web
+            dir="$(dirname "$dockerfile")"
-          docker push "$IMAGE"
+            name="$(basename "$dir")"
+            TAG="$(sha1sum "$dockerfile" | cut -d' ' -f1)"
+            IMAGE="git.octoturge.com/octo-tech/profiles-${name}:${TAG}"
+
+            echo "::group::${name}"
+            if docker manifest inspect "$IMAGE" >/dev/null 2>&1; then
+              echo "$IMAGE already in the registry (Dockerfile unchanged), skipping build."
+            elif docker build -t "$IMAGE" "$dir" && docker push "$IMAGE"; then
+              echo "Built and pushed $IMAGE"
+            else
+              echo "::warning::Failed to build/push $IMAGE - templates/$name will be skipped this run."
+              FAILED="$FAILED $name"
+            fi
+            echo "::endgroup::"
+          done
+
+          echo "failed_templates=${FAILED# }" >> "$GITHUB_OUTPUT"
 
   provision:
-    needs: build-web-image
+    # needs: build-images orders this after the image builds (so a fresh
+    # template push never points at a tag that isn't in the registry yet)
+    # without making every template's reprovisioning depend on ALL builds
+    # succeeding - if:always() overrides the default "skip if a dependency
+    # failed" behavior, since build-images only fails outright on an
+    # infra-level problem (e.g. registry login); a single template's build
+    # failure is reported via failed_templates instead and only skips that
+    # one template below.
+    needs: build-images
+    if: always()
     runs-on: ubuntu-latest
     env:
       CODER_URL: ${{ secrets.CODER_URL }}
       CODER_SESSION_TOKEN: ${{ secrets.CODER_SESSION_TOKEN }}
+      FAILED_TEMPLATES: ${{ needs.build-images.outputs.failed_templates }}
     steps:
       - name: Checkout (full history, needed to detect removed templates)
         uses: actions/checkout@v4
@@ -81,9 +115,16 @@ jobs:
         run: |
           set -e
           for dir in templates/*/; do
-            name="profiles-$(basename "$dir")"
+            name="$(basename "$dir")"
-            echo "::group::Pushing $name from $dir"
+            full="profiles-$name"
-            coder templates push "$name" -d "$dir" --yes \
+            case " $FAILED_TEMPLATES " in
+              *" $name "*)
+                echo "::warning::Skipping $full - its Docker image failed to build this run (see build-images), leaving the previous template version in place."
+                continue
+                ;;
+            esac
+            echo "::group::Pushing $full from $dir"
+            coder templates push "$full" -d "$dir" --yes \
               -m "auto-provisioned from ${GITHUB_SHA:0:12}"
             echo "::endgroup::"
           done