feat(examples/pachyderm-cd4ml): New e2e CD demo using Pachyderm

examples/pachyderm-cd4ml/.gitignore

@@ -0,0 +1,3 @@
+*.json
+*.joblib
+rbac.yaml

examples/pachyderm-cd4ml/Makefile

@@ -0,0 +1,14 @@
+build:
+	make -C RSIServing build
+	make -C RSITraining build
+	make -C RSIDeployer build
+
+push: build
+	make -C RSIServing push
+	make -C RSITraining push
+	make -C RSIDeployer push
+
+kind_load: build
+	make -C RSIServing kind_load
+	make -C RSITraining kind_load
+	make -C RSIDeployer kind_load

examples/pachyderm-cd4ml/RSIDeployer/Dockerfile

@@ -0,0 +1,4 @@
+FROM alpine/helm:3.4.2
+ADD ./charts /charts
+ADD ./deploy.sh /
+ENTRYPOINT [ "ash" ]

examples/pachyderm-cd4ml/RSIDeployer/Makefile

@@ -0,0 +1,12 @@
+VERSION=0.1.0
+IMAGE_BASE=seldonio/pachyderm-rsi-deployer
+KIND_NAME ?= kind
+
+build:
+	docker build . -t ${IMAGE_BASE}:${VERSION}
+
+push: build
+	docker push ${IMAGE_BASE}:${VERSION}
+
+kind_load: build
+	kind load -v 3 docker-image ${IMAGE_BASE}:${VERSION} --name ${KIND_NAME}

examples/pachyderm-cd4ml/RSIDeployer/charts/seldon-deployment/.helmignore

@@ -0,0 +1,23 @@
+# Patterns to ignore when building packages.
+# This supports shell glob matching, relative path matching, and
+# negation (prefixed with !). Only one pattern per line.
+.DS_Store
+# Common VCS dirs
+.git/
+.gitignore
+.bzr/
+.bzrignore
+.hg/
+.hgignore
+.svn/
+# Common backup files
+*.swp
+*.bak
+*.tmp
+*.orig
+*~
+# Various IDEs
+.project
+.idea/
+*.tmproj
+.vscode/

examples/pachyderm-cd4ml/RSIDeployer/charts/seldon-deployment/Chart.yaml

@@ -0,0 +1,6 @@
+apiVersion: v2
+name: seldon-deployment
+description: A custom helm chart for this seldon deployment
+type: application
+version: 0.1.0
+appVersion: 0.1.0

examples/pachyderm-cd4ml/RSIDeployer/charts/seldon-deployment/templates/secret.yaml

@@ -0,0 +1,17 @@
+apiVersion: v1
+kind: Secret
+metadata:
+  name: minio-credentials
+type: Opaque
+stringData:
+  AWS_ACCESS_KEY_ID: noauth
+  AWS_SECRET_ACCESS_KEY: noauth
+  AWS_ENDPOINT_URL: http://pachd.pachyderm.svc.cluster.local:600
+  USE_SSL: "false"
+---
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: rsi-deployment-sa
+secrets:
+  - name: minio-credentials

examples/pachyderm-cd4ml/RSIDeployer/charts/seldon-deployment/templates/seldon-deployment.yaml

@@ -0,0 +1,40 @@
+apiVersion: machinelearning.seldon.io/v1
+kind: SeldonDeployment
+metadata:
+  name: rsi-predictor
+spec:
+  annotations:
+    seldon.io/executor: "true"
+  name: rsi-predictor
+  predictors:
+  - componentSpecs:
+    - spec:
+        containers:
+        - image: seldonio/pachyderm-rsi-server:0.1.0
+          imagePullPolicy: IfNotPresent
+          name: rsi-predictor
+          env:
+          - name: MODEL_URI
+            value: {{ .Values.model.uri }}
+          - name: MODEL_COMMIT_HASH
+            value: {{ .Values.model.version }}
+          - name: TEMP_DIR
+            value: /mnt/models
+          envFrom:
+          - secretRef:
+              name: minio-credentials
+          volumeMounts:
+          - name: model-storage
+            mountPath: /mnt/models
+        volumes:
+        - name: model-storage
+          emptyDir: {}
+        serviceAccountName: rsi-deployment-sa
+    graph:
+      children: []
+      endpoint:
+        type: REST
+      name: rsi-predictor
+      type: MODEL
+    name: default
+    replicas: 1

examples/pachyderm-cd4ml/RSIDeployer/charts/seldon-deployment/values.yaml

@@ -0,0 +1,3 @@
+model:
+  uri: "uri_to_model"
+  version: "commit_hash"

examples/pachyderm-cd4ml/RSIDeployer/deploy.sh

@@ -0,0 +1,14 @@
+#!/bin/ash
+
+if [ $# -eq 0 ]
+  then
+    echo "USAGE: deploy.sh NAMESPACE MODEL_URI MODEL_VERSION"
+fi
+
+echo "namespace=${1} model.uri=${2} model.version=${3}"
+
+if helm --namespace=${1} status rsi ; then
+    helm --namespace=${1} upgrade rsi /charts/seldon-deployment --set model.uri=${2} --set model.version=${3}
+else
+    helm --namespace=${1} install rsi /charts/seldon-deployment --set model.uri=${2} --set model.version=${3}
+fi

examples/pachyderm-cd4ml/RSIServing/Dockerfile

@@ -0,0 +1,6 @@
+FROM python:3.8-slim
+RUN pip install pandas==1.1.4 statsmodels==0.12.1 seldon-core==1.5.0 joblib==1.0.0 minio==7.0.0
+WORKDIR /app
+ADD RSIModel.py /app/RSIModel.py
+EXPOSE 5000
+CMD exec seldon-core-microservice RSIModel --service-type MODEL --persistence 0

examples/pachyderm-cd4ml/RSIServing/Makefile

@@ -0,0 +1,12 @@
+VERSION=0.1.0
+IMAGE_BASE=seldonio/pachyderm-rsi-server
+KIND_NAME ?= kind
+
+build:
+	docker build . -t ${IMAGE_BASE}:${VERSION}
+
+push: build
+	docker push ${IMAGE_BASE}:${VERSION}
+
+kind_load: build
+	kind load -v 3 docker-image ${IMAGE_BASE}:${VERSION} --name ${KIND_NAME}

examples/pachyderm-cd4ml/RSIServing/RSIModel.py

@@ -0,0 +1,90 @@
+import logging
+import os
+from urllib.parse import urlparse
+
+import joblib
+from minio import Minio
+
+MODEL_URI = os.getenv("MODEL_URI", "")
+MODEL_COMMIT_HASH = os.getenv("MODEL_COMMIT_HASH", "")
+TEMP_DIR = os.getenv("TEMP_DIR", "/mnt/models")
+_S3_PREFIX = "s3://"
+
+logger = logging.getLogger(__name__)
+
+
+class RSIModel(object):
+    def __init__(self):
+        super().__init__()
+        self.ready = False
+        logger.info(f"Model uri: {MODEL_URI} @ ${MODEL_COMMIT_HASH}")
+        model_file = self._download_s3()
+        logger.info(f"model file: {model_file}")
+        self._model = joblib.load(model_file)
+
+    def predict(self, X, features_names):
+        """Predict!
+
+        The return type is important. It has to fit into one of three types:
+        https://docs.seldon.io/projects/seldon-core/en/v1.1.0/python/python_component.html
+
+        The statsmodel model returns a pandas Series. I want to return the index, so I reset
+        it to move it into a column, then use the to_numpy method to convert it to a valid
+        seldon return type.
+
+        Args:
+            X (int): A scalar representing the number of periods to forecast.
+
+        Returns:
+            np.ndarray: The forecast [timestamps x predictions]
+        """
+        return self._model.forecast(X).reset_index().to_numpy()
+
+    def init_metadata(self):
+        meta = {
+            "name": "RSIModel",
+            "versions": [f"{MODEL_URI}@{MODEL_COMMIT_HASH}"],
+            "platform": "seldon",
+            "inputs": [
+                {
+                    "messagetype": "ndarray",
+                    "schema": {"names": ["periods"], "shape": [1]},
+                }
+            ],
+            "outputs": [{"messagetype": "ndarray", "schema": {"shape": [2]}}],
+            "custom": {"MODEL_URI": MODEL_URI, "MODEL_COMMIT_HASH": MODEL_COMMIT_HASH},
+        }
+        return meta
+
+    # Inspired by https://github.com/SeldonIO/kfserving/blob/master/python/kfserving/kfserving/storage.py
+    def _download_s3(self) -> str:
+        client = RSIModel._create_minio_client()
+        bucket_args = MODEL_URI.replace(_S3_PREFIX, "", 1).split("/", 1)
+        bucket_name = bucket_args[0]
+        bucket_path = bucket_args[1] if len(bucket_args) > 1 else ""
+        file_name = bucket_path.split("/")[-1]
+        local_file = os.path.join(TEMP_DIR, file_name)
+        logger.info(f"Downloading {bucket_path} from {bucket_name} to {local_file}.")
+        try:
+            client.fget_object(
+                bucket_name, bucket_path, local_file, version_id=MODEL_COMMIT_HASH
+            )
+        except Exception as e:
+            raise RuntimeError(f"Failed to fetch model {MODEL_URI}.\n{e}")
+        return local_file
+
+    @staticmethod
+    def _create_minio_client():
+        # Remove possible http scheme for Minio
+        url = urlparse(os.getenv("AWS_ENDPOINT_URL", "s3.amazonaws.com"))
+        use_ssl = (
+            url.scheme == "https"
+            if url.scheme
+            else bool(os.getenv("S3_USE_HTTPS", "true"))
+        )
+        return Minio(
+            url.netloc,
+            access_key=os.getenv("AWS_ACCESS_KEY_ID", ""),
+            secret_key=os.getenv("AWS_SECRET_ACCESS_KEY", ""),
+            secure=use_ssl,
+        )

examples/pachyderm-cd4ml/RSITraining/Dockerfile

@@ -0,0 +1,4 @@
+FROM python:3.8-slim
+RUN pip install pandas==1.1.4 statsmodels==0.12.1 joblib==1.0.0
+WORKDIR /code
+ADD RSITrain.py /code/RSITrain.py

examples/pachyderm-cd4ml/RSITraining/Makefile

@@ -0,0 +1,14 @@
+VERSION=0.1.0
+IMAGE_BASE=seldonio/pachyderm-rsi-trainer
+
+KIND_NAME ?= kind
+
+
+build:
+	docker build . -t ${IMAGE_BASE}:${VERSION}
+
+push:
+	docker push ${IMAGE_BASE}:${VERSION}
+
+kind_load:
+	kind load -v 3 docker-image ${IMAGE_BASE}:${VERSION} --name ${KIND_NAME}

examples/pachyderm-cd4ml/RSITraining/RSITrain.py

@@ -0,0 +1,21 @@
+import argparse
+
+import joblib
+import pandas as pd
+from statsmodels.tsa.holtwinters import ExponentialSmoothing
+
+parser = argparse.ArgumentParser(description='Train a model for iris classification.')
+parser.add_argument('data_file_path', type=str, help='Path to csv file containing training set')
+parser.add_argument('model_path', type=str, help='Path to output joblib file')
+args = parser.parse_args()
+
+print(f"Loading data set from {args.data_file_path}")
+rsi = pd.read_csv(args.data_file_path, index_col="Date", parse_dates=True)
+
+print("Training model...")
+model = ExponentialSmoothing(rsi,trend='add').fit()
+print("Model trained!")
+
+print(f"Saving model in {args.model_path}")
+joblib.dump(model, args.model_path)
+print("Model saved!")

examples/pachyderm-cd4ml/data/rsi_after.csv

@@ -0,0 +1,83 @@
+"Date","RSI"
+"2014 JAN",86.0
+"2014 FEB",86.4
+"2014 MAR",87.5
+"2014 APR",87.2
+"2014 MAY",87.3
+"2014 JUN",88.0
+"2014 JUL",87.8
+"2014 AUG",87.7
+"2014 SEP",87.0
+"2014 OCT",87.4
+"2014 NOV",87.8
+"2014 DEC",89.0
+"2015 JAN",87.2
+"2015 FEB",87.7
+"2015 MAR",88.0
+"2015 APR",88.5
+"2015 MAY",88.5
+"2015 JUN",88.9
+"2015 JUL",88.7
+"2015 AUG",87.6
+"2015 SEP",89.2
+"2015 OCT",87.9
+"2015 NOV",88.4
+"2015 DEC",88.3
+"2016 JAN",89.8
+"2016 FEB",89.1
+"2016 MAR",88.8
+"2016 APR",89.4
+"2016 MAY",90.7
+"2016 JUN",90.0
+"2016 JUL",91.7
+"2016 AUG",91.3
+"2016 SEP",92.1
+"2016 OCT",93.9
+"2016 NOV",93.6
+"2016 DEC",93.2
+"2017 JAN",93.2
+"2017 FEB",94.3
+"2017 MAR",94.0
+"2017 APR",95.5
+"2017 MAY",94.4
+"2017 JUN",94.7
+"2017 JUL",95.1
+"2017 AUG",96.1
+"2017 SEP",96.4
+"2017 OCT",96.8
+"2017 NOV",97.7
+"2017 DEC",97.7
+"2018 JAN",97.8
+"2018 FEB",98.4
+"2018 MAR",97.0
+"2018 APR",98.2
+"2018 MAY",100.4
+"2018 JUN",100.0
+"2018 JUL",100.7
+"2018 AUG",101.4
+"2018 SEP",101.0
+"2018 OCT",100.7
+"2018 NOV",102.5
+"2018 DEC",101.9
+"2019 JAN",102.5
+"2019 FEB",103.1
+"2019 MAR",104.2
+"2019 APR",103.4
+"2019 MAY",103.2
+"2019 JUN",104.4
+"2019 JUL",105.1
+"2019 AUG",104.5
+"2019 SEP",104.0
+"2019 OCT",104.0
+"2019 NOV",103.4
+"2019 DEC",103.1
+"2020 JAN",104.5
+"2020 FEB",103.6
+"2020 MAR",97.4
+"2020 APR",79.4
+"2020 MAY",88.7
+"2020 JUN",100.9
+"2020 JUL",105.3
+"2020 AUG",106.2
+"2020 SEP",107.5
+"2020 OCT",109.0