[wip] Fix CI for publishing plugin images

.github/workflows/publish.yml

@@ -7,6 +7,7 @@ on:
     paths:
       - '.github/workflows/publish.yml'
       - 'maint_tools/build_default_image.py'
+      - 'maint_tools/build_plugin_image.py'
 
 jobs:
   flyte-pypi:
@@ -89,6 +90,17 @@ jobs:
         uses: actions/setup-python@v1
         with:
           python-version: "3.13"
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+      - name: Set up Docker Buildx
+        id: buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Login to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io/flyteorg
+          username: "${{ secrets.FLYTE_BOT_USERNAME }}"
+          password: "${{ secrets.FLYTE_BOT_PAT }}"
       - name: Install uv
         uses: astral-sh/setup-uv@v6
         id: setup-uv
@@ -97,7 +109,7 @@ jobs:
         run: |
           uv venv
           uv pip install build twine setuptools wheel
-      - name: Build and publish
+      - name: Build
         working-directory: ${{ matrix.workdir }}
         run: |
           uv run --prerelease allow python -m build --wheel --installer uv

maint_tools/build_plugin_image.py

@@ -8,11 +8,10 @@
 async def build_flyte_vllm_image(registry: str | None = None, name: str | None = None, builder: str | None = "local"):
     """
     Build the SDK default Debian-based image, optionally overriding
-    the container registry and image name.
+    the container registry name.
 
     Args:
         registry: e.g. "ghcr.io/my-org" or "123456789012.dkr.ecr.us-west-2.amazonaws.com".
-        name:     e.g. "my-flyte-image".
         builder:  e.g. "local" or "remote".
     """
     import flyteplugins.vllm
@@ -24,39 +23,45 @@ async def build_flyte_vllm_image(registry: str | None = None, name: str | None =
 async def build_flyte_sglang_image(registry: str | None = None, name: str | None = None, builder: str | None = "local"):
     """
     Build the SDK default Debian-based image, optionally overriding
-    the container registry and image name.
+    the container registry name.
     """
     import flyteplugins.sglang
 
     default_image = flyteplugins.sglang.DEFAULT_SGLANG_IMAGE.clone(registry=registry, name=name)
     await ImageBuildEngine.build(default_image, builder=builder)
 
 
-async def build_all(registry: str | None = None, name: str | None = None, builder: str | None = "local"):
-    await asyncio.gather(
-        build_flyte_vllm_image(registry=registry, name=name, builder=builder),
-        build_flyte_sglang_image(registry=registry, name=name, builder=builder),
-    )
-
-
 if __name__ == "__main__":
+    import os
+
     parser = argparse.ArgumentParser(description="Build the default Flyte image.")
     parser.add_argument("--registry", help="Docker registry to push to")
-    parser.add_argument("--name", help="Custom image name (without tag)")
-    parser.add_argument("--type", choices=["vllm", "sglang", "all"], help="Type of image to build")
+    parser.add_argument("--type", choices=["vllm", "sglang"], help="Type of image to build")
     parser.add_argument("--builder", choices=["local", "remote"], default="local", help="Image builder to use")
 
     args = parser.parse_args()
-    # can remove this and only specify one in the future
-    assert (args.registry and args.name) or (not args.registry and not args.name)
 
-    flyte.init_from_config()
+    if os.getenv("GITHUB_ACTIONS", "") == "true":
+        flyte.init(
+            endpoint=os.getenv("FLYTE_ENDPOINT", "dns:///playground.canary.unionai.cloud"),
+            auth_type="ClientSecret",
+            client_id="flyte-sdk-ci",
+            client_credentials_secret=os.getenv("FLYTE_SDK_CI_TOKEN"),
+            insecure=False,
+            image_builder="remote",
+            project=os.getenv("FLYTE_PROJECT", "flyte-sdk"),
+            domain=os.getenv("FLYTE_DOMAIN", "development"),
+        )
+        builder = "remote"
+    else:
+        flyte.init_from_config()
+        builder = args.builder
+
     if args.type == "vllm":
         print("Building vllm image...")
-        asyncio.run(build_flyte_vllm_image(registry=args.registry, name=args.name, builder=args.builder))
+        asyncio.run(build_flyte_vllm_image(registry=args.registry, builder=builder))
     elif args.type == "sglang":
         print("Building sglang image...")
-        asyncio.run(build_flyte_sglang_image(registry=args.registry, name=args.name, builder=args.builder))
+        asyncio.run(build_flyte_sglang_image(registry=args.registry, builder=builder))
     else:
-        print("Building all images...")
-        asyncio.run(build_all(registry=args.registry, name=args.name, builder=args.builder))
+        raise ValueError(f"Invalid type: {args.type}. Valid types are: [vllm, sglang].")

plugins/sglang/src/flyteplugins/sglang/_app_environment.py

@@ -14,7 +14,7 @@
 from flyteplugins.sglang._constants import SGLANG_MIN_VERSION_STR
 
 DEFAULT_SGLANG_IMAGE = (
-    flyte.Image.from_debian_base(name="sglang-app-image")
+    flyte.Image.from_debian_base(name="sglang-app-image", python_version=(3, 13))
     # install system dependencies, including CUDA toolkit, which is needed by sglang for compiling the model
     # and rust and cargo for installing sglang
     .with_apt_packages("libnuma-dev", "wget", "curl", "openssl", "pkg-config", "libssl-dev", "build-essential")

plugins/vllm/src/flyteplugins/vllm/_app_environment.py

@@ -14,7 +14,7 @@
 from flyteplugins.vllm._constants import VLLM_MIN_VERSION_STR
 
 DEFAULT_VLLM_IMAGE = (
-    flyte.Image.from_debian_base(name="vllm-app-image")
+    flyte.Image.from_debian_base(name="vllm-app-image", python_version=(3, 13))
     # install flashinfer and vllm
     .with_pip_packages("flashinfer-python", "flashinfer-cubin")
     .with_pip_packages("flashinfer-jit-cache", index_url="https://flashinfer.ai/whl/cu129")