Create ensembler web service

.github/workflows/batch-ensembler.yaml → .github/workflows/pyfunc-ensembler-job.yaml

@@ -1,25 +1,25 @@
-name: engines/batch-ensembler
+name: engines/pyfunc-ensembler-job
 
 on:
   # Automatically run CI on Release and Pre-Release tags and main branch
   # (only if there are changes to relevant paths)
   push:
     tags:
-      - "batch-ensembler/v[0-9]+.[0-9]+.[0-9]+*"
+      - "pyfunc-ensembler-job/v[0-9]+.[0-9]+.[0-9]+*"
     branches:
       - main
     paths:
-      - ".github/workflows/batch-ensembler.yaml"
-      - "engines/batch-ensembler/**"
+      - ".github/workflows/pyfunc-ensembler-job.yaml"
+      - "engines/pyfunc-ensembler-job/**"
       - "sdk/**"
 
   # Automatically run CI on branches, that have active PR opened
   pull_request:
     branches:
       - main
     paths:
-      - ".github/workflows/batch-ensembler.yaml"
-      - "engines/batch-ensembler/**"
+      - ".github/workflows/pyfunc-ensembler-job.yaml"
+      - "engines/pyfunc-ensembler-job/**"
       - "sdk/**"
 
   # To make it possible to trigger e2e CI workflow for any arbitrary git ref
@@ -50,13 +50,13 @@ jobs:
       - name: Cache Conda environment
         uses: actions/cache@v2
         with:
-          path: engines/batch-ensembler/env
+          path: engines/pyfunc-ensembler-job/env
           key: |
-            conda-${{ hashFiles('engines/batch-ensembler/environment.yaml') }}-${{ hashFiles('engines/batch-ensembler/requirements.txt') }}-${{ hashFiles('engines/batch-ensembler/requirements.dev.txt') }}
+            conda-${{ hashFiles('engines/pyfunc-ensembler-job/environment.yaml') }}-${{ hashFiles('engines/pyfunc-ensembler-job/requirements.txt') }}-${{ hashFiles('engines/pyfunc-ensembler-job/requirements.dev.txt') }}
           restore-keys: conda-
 
       - name: Run Tests
-        working-directory: engines/batch-ensembler
+        working-directory: engines/pyfunc-ensembler-job
         run: |
           make setup
           make test
@@ -70,7 +70,7 @@ jobs:
       - id: release-rules
         uses: ./.github/actions/release-rules
         with:
-          prefix: batch-ensembler/
+          prefix: pyfunc-ensembler-job/
 
   publish:
     # Automatically publish release and pre-release artifacts.
@@ -103,13 +103,13 @@ jobs:
 
       - name: Build Docker Image
         id: build
-        working-directory: engines/batch-ensembler
+        working-directory: engines/pyfunc-ensembler-job
         env:
           DOCKER_REGISTRY: ghcr.io/${{ github.repository }}
         run: |
           set -o pipefail
           make build-image | tee output.log
-          echo "::set-output name=ensembler-image::$(sed -n 's%Building docker image: \(.*\)%\1%p' output.log)"
+          echo "::set-output name=pyfunc-ensembler-job::$(sed -n 's%Building docker image: \(.*\)%\1%p' output.log)"
 
-      - name: Publish Batch Ensembler Docker Image
-        run: docker push ${{ steps.build.outputs.ensembler-image }}
+      - name: Publish Pyfunc Ensembler Job Docker Image
+        run: docker push ${{ steps.build.outputs.pyfunc-ensembler-job }}

.github/workflows/pyfunc-ensembler-service.yaml

@@ -0,0 +1,109 @@
+name: engines/pyfunc-ensembler-service
+
+on:
+  # Automatically run CI on Release and Pre-Release tags and main branch
+  # (only if there are changes to relevant paths)
+  push:
+    tags:
+      - "pyfunc-ensembler-service/v[0-9]+.[0-9]+.[0-9]+*"
+    branches:
+      - main
+    paths:
+      - ".github/workflows/pyfunc-ensembler-service.yaml"
+      - "engines/pyfunc-ensembler-service/**"
+      - "sdk/**"
+
+  # Automatically run CI on branches, that have active PR opened
+  pull_request:
+    branches:
+      - main
+    paths:
+      - ".github/workflows/pyfunc-ensembler-service.yaml"
+      - "engines/pyfunc-ensembler-service/**"
+      - "sdk/**"
+
+  # To make it possible to trigger e2e CI workflow for any arbitrary git ref
+  workflow_dispatch:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Setup Python
+        uses: actions/setup-python@v2
+        with:
+          python-version: 3.8
+
+      - name: Setup Conda
+        uses: conda-incubator/setup-miniconda@v2
+        with:
+          auto-update-conda: true
+
+      - name: Cache Conda environment
+        uses: actions/cache@v2
+        with:
+          path: engines/pyfunc-ensembler-service/env
+          key: |
+            conda-${{ hashFiles('engines/pyfunc-ensembler-service/environment.yaml') }}-${{ hashFiles('engines/pyfunc-ensembler-service/requirements.txt') }}-${{ hashFiles('engines/pyfunc-ensembler-service/requirements.dev.txt') }}
+          restore-keys: conda-
+
+      - name: Run Tests
+        working-directory: engines/pyfunc-ensembler-service
+        run: |
+          make setup
+          make test
+
+  release-rules:
+    runs-on: ubuntu-latest
+    outputs:
+      release-type: ${{ steps.release-rules.outputs.release-type }}
+    steps:
+      - uses: actions/checkout@v2
+      - id: release-rules
+        uses: ./.github/actions/release-rules
+        with:
+          prefix: pyfunc-ensembler-service/
+
+  publish:
+    # Automatically publish release and pre-release artifacts.
+    #
+    # As for dev releases, make it possible to publish artifacts
+    # manually by approving 'deployment' in the 'manual' environment.
+    #
+    # Dev build can be released either from the 'main' branch or
+    # by running this workflow manually with `workflow_dispatch` event.
+    if: >-
+      contains('release,pre-release', needs.release-rules.outputs.release-type)
+        || ( github.event_name != 'pull_request' )
+        || ( github.event.pull_request.head.repo.full_name == github.repository )
+    environment: ${{ needs.release-rules.outputs.release-type == 'dev' && 'manual' || '' }}
+    runs-on: ubuntu-latest
+    needs:
+      - release-rules
+      - test
+    steps:
+      - uses: actions/checkout@v2
+        with:
+          fetch-depth: 0
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v1
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Build Docker Image
+        id: build
+        working-directory: engines/pyfunc-ensembler-service
+        env:
+          DOCKER_REGISTRY: ghcr.io/${{ github.repository }}
+        run: |
+          set -o pipefail
+          make build-image | tee output.log
+          echo "::set-output name=pyfunc-ensembler-service-image::$(sed -n 's%Building docker image: \(.*\)%\1%p' output.log)"
+
+      - name: Publish Pyfunc Ensembler Service Docker Image
+        run: docker push ${{ steps.build.outputs.pyfunc-ensembler-service-image }}

.github/workflows/turing.yaml

@@ -10,9 +10,11 @@ on:
       - main
     paths-ignore:
       - "docs/**"
-      - "engines/batch-ensembler/**"
+      - "engines/pyfunc-ensembler-job/**"
+      - "engines/pyfunc-ensembler-service/**"
       - "sdk/**"
-      - ".github/workflows/batch-ensembler.yaml"
+      - ".github/workflows/pyfunc-ensembler-job.yaml"
+      - ".github/workflows/pyfunc-ensembler-service.yaml"
       - ".github/workflows/sdk.yaml"
       - ".github/workflows/helm-chart.yaml"
       - ".github/workflows/cluster-init.yaml"
@@ -23,9 +25,11 @@ on:
       - main
     paths-ignore:
       - "docs/**"
-      - "engines/batch-ensembler/**"
+      - "engines/pyfunc-ensembler-job/**"
+      - "engines/pyfunc-ensembler-service/**"
       - "sdk/**"
-      - ".github/workflows/batch-ensembler.yaml"
+      - ".github/workflows/pyfunc-ensembler-job.yaml"
+      - ".github/workflows/pyfunc-ensembler-service.yaml"
       - ".github/workflows/sdk.yaml"
       - ".github/workflows/helm-chart.yaml"
 

README.md

@@ -180,10 +180,10 @@ BatchEnsemblingConfig:
     BuildNamespace: default
     BuildTimeoutDuration: 20m
     DestinationRegistry: ghcr.io
-    BaseImageRef: ghcr.io/gojek/turing/batch-ensembler:latest
+    BaseImageRef: ghcr.io/gojek/turing/pyfunc-ensembler-job:latest
     KanikoConfig:
       BuildContextURI: git://github.com/gojek/turing.git#refs/heads/main
-      DockerfileFilePath: engines/batch-ensembler/app.Dockerfile
+      DockerfileFilePath: engines/pyfunc-ensembler-job/app.Dockerfile
       Image: gcr.io/kaniko-project/executor
       ImageVersion: v1.6.0
       ResourceRequestsLimits:

api/config-dev.yaml

@@ -19,10 +19,10 @@ BatchEnsemblingConfig:
     BuildNamespace: default
     BuildTimeoutDuration: 20m
     DestinationRegistry: ghcr.io
-    BaseImageRef: ghcr.io/gojek/turing/batch-ensembler:latest
+    BaseImageRef: ghcr.io/gojek/turing/pyfunc-ensembler-job:latest
     KanikoConfig:
       BuildContextURI: git://github.com/gojek/turing.git#refs/heads/main
-      DockerfileFilePath: engines/batch-ensembler/app.Dockerfile
+      DockerfileFilePath: engines/pyfunc-ensembler-job/app.Dockerfile
       Image: gcr.io/kaniko-project/executor
       ImageVersion: v1.6.0
       ResourceRequestsLimits:

api/turing/config/example.yaml

@@ -42,10 +42,10 @@ BatchEnsemblingConfig:
     BuildNamespace: default
     BuildTimeoutDuration: 20m
     DestinationRegistry: ghcr.io
-    BaseImageRef: ghcr.io/gojek/turing/batch-ensembler:latest
+    BaseImageRef: ghcr.io/gojek/turing/pyfunc-ensembler-job:latest
     KanikoConfig:
       BuildContextURI: git://github.com/gojek/turing.git#refs/heads/main
-      DockerfileFilePath: engines/batch-ensembler/app.Dockerfile
+      DockerfileFilePath: engines/pyfunc-ensembler-job/app.Dockerfile
       Image: gcr.io/kaniko-project/executor
       ImageVersion: v1.6.0
       ResourceRequestsLimits:

engines/batch-ensembler/Makefile → engines/pyfunc-ensembler-job/Makefile

@@ -1,6 +1,6 @@
 SHELL := /bin/bash
 
-APP_NAME := batch-ensembler
+APP_NAME := pyfunc-ensembler-job
 CONDA_ENV_NAME ?= $(APP_NAME)
 ACTIVATE_ENV = source $$(conda info --base)/etc/profile.d/conda.sh ; conda activate ./env/$(CONDA_ENV_NAME)
 
@@ -39,4 +39,4 @@ build-image: version
 .PHONY: version
 version:
 	$(eval VERSION=$(if $(OVERWRITE_VERSION),$(OVERWRITE_VERSION),v$(shell ../../scripts/vertagen/vertagen.sh -p ${APP_NAME}/)))
-	@echo "turing-batch-ensembler version:" $(VERSION)
+	@echo "turing-pyfunc-ensembler-job version:" $(VERSION)

engines/batch-ensembler/environment.yaml → engines/pyfunc-ensembler-job/environment.yaml

@@ -1,4 +1,4 @@
-name: batch-ensembler
+name: pyfunc-ensembler-job
 dependencies:
   - python=3.8
   - pip=21.0.1

engines/batch-ensembler/setup.py → engines/pyfunc-ensembler-job/setup.py

@@ -18,7 +18,7 @@
     ]
 
 setuptools.setup(
-    name='batch-ensembler',
+    name='pyfunc-ensembler-job',
     packages=setuptools.find_packages(),
     install_requires=requirements,
     dev_requirements=dev_requirements,

engines/pyfunc-ensembler-service/.dockerignore

@@ -0,0 +1,8 @@
+.gitignore
+.dockerignore
+
+env/
+tests/
+
+.mypy_cache/
+.pytest_cache/

engines/pyfunc-ensembler-service/.gitignore

@@ -0,0 +1,6 @@
+env/
+.coverage
+**/mlruns/
+**/__pycache__
+
+ensembler/*

engines/pyfunc-ensembler-service/Dockerfile

@@ -0,0 +1,16 @@
+FROM continuumio/miniconda3 AS builder
+
+RUN wget -qO- https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-sdk-367.0.0-linux-x86_64.tar.gz  | tar xzf -
+ENV PATH=$PATH:/google-cloud-sdk/bin
+ENV CONDA_ENV_NAME=${CONDA_ENV_NAME}
+ENV APP_NAME=${APP_NAME}
+
+COPY . .
+COPY ./temp-deps/sdk ./../../sdk
+
+RUN conda env create -f ./environment.yaml &&  \
+    conda env update --name ${CONDA_ENV_NAME} --file /ensembler/conda.yaml && \
+    rm -rf /root/.cache
+
+# Install conda-pack:
+RUN conda install -c conda-forge conda-pack

engines/pyfunc-ensembler-service/Makefile

@@ -0,0 +1,33 @@
+SHELL := /bin/bash
+
+APP_NAME := pyfunc-ensembler-service
+CONDA_ENV_NAME ?= $(APP_NAME)
+ACTIVATE_ENV = source $$(conda info --base)/etc/profile.d/conda.sh ; conda activate $(CONDA_ENV_NAME)
+
+.PHONY: setup
+setup: $(CONDA_ENV_NAME)
+$(CONDA_ENV_NAME):
+	@conda env update -f environment.yaml --prune
+	$(ACTIVATE_ENV) && pip install -r requirements.dev.txt
+
+.PHONY: test
+test:
+	@$(ACTIVATE_ENV) && \
+		python -m pytest \
+		--cov=pyfunc_ensembler_runner \
+		--cov-report term-missing \
+		-W ignore
+
+.PHONY: build-image
+build-image: version
+	@mkdir -p temp-deps
+	@cp -r ../../sdk temp-deps/
+	@$(eval IMAGE_TAG = $(if $(DOCKER_REGISTRY),$(DOCKER_REGISTRY)/,)${APP_NAME}:${VERSION})
+	@echo "Building docker image: ${IMAGE_TAG}"
+	@docker build . --tag ${IMAGE_TAG}
+	@rm -rf temp-deps
+
+.PHONY: version
+version:
+	$(eval VERSION=$(if $(OVERWRITE_VERSION),$(OVERWRITE_VERSION),v$(shell ../../scripts/vertagen/vertagen.sh -p ${APP_NAME}/)))
+	@echo "turing-pyfunc-ensembler-service version:" $(VERSION)

engines/pyfunc-ensembler-service/README.md

@@ -0,0 +1,27 @@
+# PyFuncEnsembler Server for Real-Time Experiments
+
+PyFuncEnsemblerRunner is a tool for deploying user-defined ensemblers (for use with Turing routers), written in 
+MLflow's `pyfunc` flavour.
+
+## Usage
+To run the ensembler as a webservice:
+```bash
+python -m pyfunc_ensembler_runner --mlflow_ensembler_dir $ENSEMBLER_DIR [-l {DEBUG,INFO,WARNING,ERROR,CRITICAL}]
+
+arguments: 
+  --mlflow_ensembler_dir <path/to/ensembler/dir/>       Path to the ensembler folder containing the mlflow files
+  --log-level <DEBUG||INFO||WARNING||ERROR||CRITICAL>   Set the logging level
+  -h, --help                                            Show this help message and exit
+```
+
+## Docker Image Building
+
+To create a docker image locally, you'll need to first download the model artifacts from the MLflow's model registry:
+```bash
+gsutil cp -r gs://[bucket-name]/mlflow/[project_id]/[run_id]/artifacts/ensembler .
+```
+
+To build the docker image, run the following:
+```bash
+make build-image
+```