build image

Author: chrisdengso

.github/workflows/build-image.yml

@@ -0,0 +1,55 @@
+name: build-image
+
+# Build the service container image and push it to GHCR. The k8s-manifests
+# repo pins the Deployment to an immutable sha-<commit> tag produced here
+# (Argo Image Updater bumps kustomization newTag), so a deploy is a manifest
+# bump in that repo — never a moving :latest in-cluster.
+#
+# Builds the CPU/ONNX image for now. To move to GPU, switch `file` to
+# dockerfiles/Dockerfile_trt — note the nvcr TensorRT base is ~10GB, which
+# needs a free-disk-space step and overflows the 10GB GHA layer cache.
+on:
+  push:
+    branches: [master]
+    tags: ['v*']
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  packages: write
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: docker/setup-buildx-action@v3
+
+      - name: Log in to GHCR
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ghcr.io/moderation-api/insightface-rest
+          tags: |
+            type=sha,format=long
+            type=ref,event=tag
+            type=raw,value=latest,enable={{is_default_branch}}
+
+      - name: Build and push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: dockerfiles/Dockerfile_cpu
+          platforms: linux/amd64
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

dockerfiles/Dockerfile_cpu

@@ -25,6 +25,13 @@ WORKDIR /app
 COPY if_rest /app/if_rest
 COPY entrypoint.sh /app/entrypoint.sh
 
+# Bake in the model manifest and the /health self-check images so the image
+# is self-contained for k8s. Compose bind-mounts ../models over /models, so
+# this stays compatible with local development; in-cluster, only the model
+# cache dirs are volume-mounted (PVC) and models.json is read from the image.
+COPY models/models.json /models/models.json
+COPY misc/test_images /images/test_images
+
 
 ENTRYPOINT [ "bash" ]
 CMD ["entrypoint.sh"]

dockerfiles/Dockerfile_trt

@@ -24,5 +24,13 @@ WORKDIR /app
 COPY if_rest /app/if_rest
 COPY entrypoint.sh /app/entrypoint.sh
 
+# Bake in the model manifest and the /health self-check images so the image
+# is self-contained for k8s. Compose bind-mounts ../models over /models, so
+# this stays compatible with local development; in-cluster, only the onnx/
+# and trt-engines/ cache dirs are volume-mounted (PVC) and models.json is
+# read from the image.
+COPY models/models.json /models/models.json
+COPY misc/test_images /images/test_images
+
 ENTRYPOINT [ "bash" ]
 CMD ["entrypoint.sh"]