feat: Custom Docker image for Bytewax batch materialization (#3099)

Dockerfile and instructions for building a custom Bytewax image.

Signed-off-by: Dan Herrera <[email protected]>

Signed-off-by: Dan Herrera <[email protected]>

docs/reference/batch-materialization/bytewax.md

@@ -55,5 +55,20 @@ batch_engine:
 
 The `namespace` configuration directive specifies which Kubernetes [namespace](https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/) jobs, services and configuration maps will be created in.
 
-The `image` parameter specifies which container image to use when running the materialization job. To create a custom image based on this container, please see the [GitHub repository](https://github.com/bytewax/bytewax-feast) for this image.
+#### Building a custom Bytewax Docker image
+
+The `image` configuration directive specifies which container image to use when running the materialization job. To create a custom image based on this container, run the following command:
+
+``` shell
+DOCKER_BUILDKIT=1 docker build . -f ./sdk/python/feast/infra/materialization/contrib/bytewax/Dockerfile -t <image tag>
+```
+
+Once that image is built and pushed to a registry, it can be specified as a part of the batch engine configuration:
+
+``` shell
+batch_engine:
+  type: bytewax
+  namespace: bytewax
+  image: <image tag>
+```
 

sdk/python/feast/infra/materialization/contrib/bytewax/Dockerfile

@@ -0,0 +1,29 @@
+FROM python:3.9-slim-bullseye AS build
+
+RUN apt-get update && \
+    apt-get install --no-install-suggests --no-install-recommends --yes git
+
+WORKDIR /bytewax
+
+# Copy dataflow code
+COPY sdk/python/feast/infra/materialization/contrib/bytewax/bytewax_materialization_dataflow.py /bytewax
+COPY sdk/python/feast/infra/materialization/contrib/bytewax/dataflow.py /bytewax
+
+# Copy entrypoint
+COPY sdk/python/feast/infra/materialization/contrib/bytewax/entrypoint.sh /bytewax
+
+# Copy necessary parts of the Feast codebase
+COPY sdk/python sdk/python
+COPY protos protos
+COPY go go
+COPY setup.py setup.py
+COPY pyproject.toml pyproject.toml
+COPY README.md README.md
+
+# Install Feast for AWS with Bytewax dependencies
+# We need this mount thingy because setuptools_scm needs access to the
+# git dir to infer the version of feast we're installing.
+# https://github.com/pypa/setuptools_scm#usage-from-docker
+# I think it also assumes that this dockerfile is being built from the root of the directory.
+RUN --mount=source=.git,target=.git,type=bind pip3 install --no-cache-dir -e '.[aws,gcp,bytewax]'
+

sdk/python/feast/infra/materialization/contrib/bytewax/bytewax_materialization_engine.py

@@ -20,7 +20,7 @@
 )
 from feast.infra.offline_stores.offline_store import OfflineStore
 from feast.infra.online_stores.online_store import OnlineStore
-from feast.registry import BaseRegistry
+from feast.infra.registry.base_registry import BaseRegistry
 from feast.repo_config import FeastConfigBaseModel
 from feast.stream_feature_view import StreamFeatureView
 from feast.utils import _get_column_names
@@ -341,7 +341,7 @@ def _create_job_definition(self, job_id, namespace, pods, env):
                             {
                                 "command": ["sh", "-c", "sh ./entrypoint.sh"],
                                 "env": job_env,
-                                "image": "bytewax/bytewax-feast:latest",
+                                "image": self.batch_engine_config.image,
                                 "imagePullPolicy": "Always",
                                 "name": "process",
                                 "ports": [

sdk/python/feast/infra/materialization/contrib/bytewax/dataflow.py

@@ -0,0 +1,22 @@
+import yaml
+
+from feast import FeatureStore, RepoConfig
+from feast.infra.materialization.contrib.bytewax.bytewax_materialization_dataflow import (
+    BytewaxMaterializationDataflow,
+)
+
+if __name__ == "__main__":
+    with open("/var/feast/feature_store.yaml") as f:
+        feast_config = yaml.safe_load(f)
+
+        with open("/var/feast/bytewax_materialization_config.yaml") as b:
+            bytewax_config = yaml.safe_load(b)
+
+            config = RepoConfig(**feast_config)
+            store = FeatureStore(config=config)
+
+            job = BytewaxMaterializationDataflow(
+                config,
+                store.get_feature_view(bytewax_config["feature_view"]),
+                bytewax_config["paths"],
+            )

sdk/python/feast/infra/materialization/contrib/bytewax/entrypoint.sh

@@ -0,0 +1,4 @@
+#!/bin/sh
+
+cd /bytewax
+python dataflow.py