Add EMR CI/CD entrypoint script

infra/scripts/codebuild-entrypoint.sh

@@ -0,0 +1,157 @@
+#!/bin/bash
+
+set -euo pipefail
+
+STEP_BREADCRUMB='~~~~~~~~'
+SECONDS=0
+TIMEFORMAT="${STEP_BREADCRUMB} took %R seconds"
+
+function maybe_build_push_docker {
+    # Build and push docker image, tagged with SHA tag, if it doesn't exist already.
+    NAME=$1
+    TARGET=$NAME-docker
+    SUFFIX=feast-$NAME
+
+    if ! aws ecr describe-images --repository-name "feast-ci/feast/$SUFFIX" "--image-ids=imageTag=${GIT_TAG}" >/dev/null ; then
+        make "build-$TARGET" "push-$TARGET" REGISTRY="${DOCKER_REPOSITORY}" VERSION="${GIT_TAG}"
+    else
+        echo "Image ${DOCKER_REPOSITORY}/$SUFFIX:$GIT_TAG already exists, skipping docker build"
+    fi
+}
+
+source infra/scripts/k8s-common-functions.sh
+
+GIT_TAG=${CODEBUILD_RESOLVED_SOURCE_VERSION}
+
+echo "########## Starting stage $STAGE for ${CODEBUILD_SOURCE_REPO_URL} ${GIT_TAG} ###########"
+
+# This seems to make builds a bit faster.
+export DOCKER_BUILDKIT=1
+
+# Workaround for COPY command in core docker image that pulls local maven repo into the image
+# itself.
+mkdir .m2 2>/dev/null || true
+mkdir deps/feast/.m2 2>/dev/null || true
+
+# Log into k8s.
+echo "${STEP_BREADCRUMB} Updating kubeconfig"
+aws eks update-kubeconfig --name "$EKS_CLUSTER_NAME"
+
+# chmod kubeconfig so it doesn't complain all the time
+chmod 755 ~/.kube/config
+
+# Sanity check that kubectl is working.
+echo "${STEP_BREADCRUMB} k8s sanity check"
+kubectl get pods
+
+case $STAGE in
+    core-docker)
+        maybe_build_push_docker core
+        ;;
+    serving-docker)
+        maybe_build_push_docker serving
+        ;;
+    jupyter-docker)
+        maybe_build_push_docker jupyter
+        ;;
+    jobservice-docker)
+        maybe_build_push_docker jobservice
+        ;;
+    ci-docker)
+        maybe_build_push_docker ci
+        ;;
+    e2e-test-emr)
+        # EMR test - runs in default namespace.
+
+        # Copy cluster config template generated for us by terraform.
+        aws s3 cp "${EMR_TEMPLATE_YML}" emr_cluster.yaml
+
+        # Delete old helm release and PVCs
+        k8s_cleanup cicd default
+
+        # Create cluster OR get existing EMR cluster id. In the latter case, clean up any steps
+        # already running there from previous test runs.
+        echo "${STEP_BREADCRUMB} Creating EMR cluster, this can take up 10 minutes."
+        CLUSTER_ID=$(time emr_cluster.py --template emr_cluster.yaml ensure --cleanup)
+
+        # Get (any) node IP. EMR will use this to connect to Kafka and Redis. We make them
+        # available to the EMR job by exposing them as NodePort services.
+        NODE_IP=$(kubectl get nodes -o custom-columns=Name:.metadata.name | tail -n1)
+
+        # Helm install everything.
+        #
+        # This may occasionally run into "provided port is already allocated" error due to
+        # https://github.com/kubernetes/kubernetes/issues/85894
+        helm_install cicd "$DOCKER_REPOSITORY" "$GIT_TAG" default \
+            --set "redis.master.service.type=NodePort" \
+            --set "redis.master.service.nodePort=32379" \
+            --set "kafka.externalAccess.service.type=NodePort" \
+            --set "kafka.externalAccess.enabled=true" \
+            --set "kafka.externalAccess.service.nodePorts[0]=30092" \
+            --set "kafka.externalAccess.service.domain=${NODE_IP}" \
+            --set "kafka.service.externalPort=30094"
+
+        # Run the test suite as a one-off pod. We could also run it here, in the codebuild container
+        # itself, but that'd require more networking setup to make feast services available
+        # outside k8s cluster.
+        kubectl delete pod ci-test-runner 2>/dev/null || true
+
+        echo "${STEP_BREADCRUMB} Running the test suite"
+        time kubectl run --rm -i ci-test-runner  \
+            --restart=Never \
+            --image="${DOCKER_REPOSITORY}/feast-ci:${GIT_TAG}" \
+            --env="CLUSTER_ID=$CLUSTER_ID" \
+            --env="STAGING_PATH=$STAGING_PATH" \
+            --env="NODE_IP=$NODE_IP" \
+            --  \
+            bash -c "mkdir src && cd src && git clone $CODEBUILD_SOURCE_REPO_URL && cd feast* && git config remote.origin.fetch '+refs/pull/*:refs/remotes/origin/pull/*' && git fetch -q && git checkout $CODEBUILD_RESOLVED_SOURCE_VERSION && ./infra/scripts/setup-e2e-env-aws.sh && ./infra/scripts/test-end-to-end-aws.sh"
+
+        ;;
+    e2e-test-sparkop)
+        # spark k8s test - runs in sparkop namespace (so it doesn't interfere with a concurrently
+        # running EMR test).
+        NAMESPACE=sparkop
+        RELEASE=sparkop
+
+        # Clean up old release
+        k8s_cleanup "$RELEASE" "$NAMESPACE"
+
+        # Helm install everything in a namespace
+        helm_install "$RELEASE" "${DOCKER_REPOSITORY}" "${GIT_TAG}" "$NAMESPACE"
+
+        # Delete old test running pod if it exists
+        kubectl delete pod -n "$NAMESPACE" ci-test-runner 2>/dev/null || true
+
+        # Delete all sparkapplication resources that may be left over from the previous test runs.
+        kubectl delete sparkapplication --all -n "$NAMESPACE" || true
+
+        # Make sure the test pod has permissions to create sparkapplication resources
+        setup_sparkop_role
+
+        # Run the test suite as a one-off pod.
+        echo "${STEP_BREADCRUMB} Running the test suite"
+        if ! time kubectl run --rm -n "$NAMESPACE" -i ci-test-runner  \
+            --restart=Never \
+            --image="${DOCKER_REPOSITORY}/feast-ci:${GIT_TAG}" \
+            --env="STAGING_PATH=$STAGING_PATH" \
+            --  \
+            bash -c "mkdir src && cd src && git clone $CODEBUILD_SOURCE_REPO_URL && cd feast* && git config remote.origin.fetch '+refs/pull/*:refs/remotes/origin/pull/*' && git fetch -q && git checkout $CODEBUILD_RESOLVED_SOURCE_VERSION && ./infra/scripts/setup-e2e-env-sparkop.sh && ./infra/scripts/test-end-to-end-sparkop.sh" ; then
+
+            readarray -t CRASHED_PODS < <(kubectl get pods --no-headers=true --namespace sparkop | grep Error | awk '{ print $1 }')
+
+            for POD in "${CRASHED_PODS[@]}"; do
+                echo "Logs from crashed pod $POD:"
+                kubectl logs --namespace sparkop "$POD"
+            done
+        fi
+
+        ;;
+    cleanup)
+        emr_cluster.py --template emr_cluster.yaml destroy
+        ;;
+    *)
+        echo "Unknown stage $STAGE"
+        ;;
+esac
+
+echo "########## Stage $STAGE took $SECONDS seconds ###########"

infra/scripts/k8s-common-functions.sh

@@ -56,7 +56,7 @@ function helm_install {
     # has some issues with unbound PVCs (that cause kubectl delete pvc to hang).
     echo "${STEP_BREADCRUMB:-} Helm installing feast"
 
-    if ! time helm install --wait "$RELEASE" ./infra/charts/feast \
+    if ! time helm install --wait "$RELEASE" "${HELM_CHART_LOCATION:-./infra/charts/feast}" \
         --timeout 15m \
         --set "feast-jupyter.image.repository=${DOCKER_REPOSITORY}/feast-jupyter" \
         --set "feast-jupyter.image.tag=${GIT_TAG}" \