Patch release for v1.0.1 (#55)

* Fixing README (#45) * Updated License Badge Colors/Logo (#46) * Allow custom subnets in canaries (#47) * Allow custom subnets in canaries * Renamed canary EKS cluster * Build integration test container programmatically (#48) * Added script to deploy new integration container * Add AWSCLI to alpine container * Fixed incorrect script path * Modified AWSCLI installation * Start docker daemon * Removed sudo * Added docker daemon nohup * Move into tests to build * Added comments and documentation references * Float release semver up to major and minor tags (#50) * Adding non-ephemeral canary support (#51) * Fixing HPO/BT deletion resource leak when SageMaker throttles Describe (#52) * Fixing HPOJob leak when SageMaker throttles DescribeHPO requests * Fixing BatchTransformJob leak when SageMaker throttles DescribeBatchTransformJob requests * Do not delete non-ephemeral cluster (#54) * Push smlogs binaries with tags (#53) * Added tagged prefix binaries * Added full variables path * Proper printf format * Move import before logging * Renamed deployment_constants

Author: goswamig

README.md

@@ -1,10 +1,10 @@
 # Amazon SageMaker Operators for Kubernetes
-![GitHub release (latest SemVer)](https://img.shields.io/github/v/release/aws/amazon-sagemaker-operator-for-k8s?sort=semver)
-[![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](http://www.apache.org/licenses/LICENSE-2.0)
-![GitHub go.mod Go version](https://img.shields.io/github/go-mod/go-version/aws/amazon-sagemaker-operator-for-k8s)
+![GitHub release (latest SemVer)](https://img.shields.io/github/v/release/aws/amazon-sagemaker-operator-for-k8s?sort=semver&logo=amazon-aws&color=232F3E)
+[![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg?color=success)](http://www.apache.org/licenses/LICENSE-2.0)
+![GitHub go.mod Go version](https://img.shields.io/github/go-mod/go-version/aws/amazon-sagemaker-operator-for-k8s?color=69D7E5)
 
 ## Introduction
-Amazon SageMaker Operators for Kubernetes are operators that can be used to train machine learning models, optimize hyperparameters for a given model, run batch transform jobs over existing models, and set up inference endpoints. With these operators, users can manage their jobs in Amazon SageMaker from their Kubernetes cluster.
+Amazon SageMaker Operators for Kubernetes are operators that can be used to train machine learning models, optimize hyperparameters for a given model, run batch transform jobs over existing models, and set up inference endpoints. With these operators, users can manage their jobs in Amazon SageMaker from their Kubernetes cluster in Amazon Elastic Kubernetes Service [EKS](http://aws.amazon.com/eks).
 
 ## Usage
 

codebuild/build_integration_container.yaml

@@ -0,0 +1,18 @@
+# This CodeBuild project is run using the docker:stable-dind container
+# Docker daemon start-up script was taken from the following URL:
+# https://docs.aws.amazon.com/codebuild/latest/userguide/sample-docker-custom-image.html
+
+version: 0.2 
+phases:
+ install:
+ commands:
+ - nohup /usr/local/bin/dockerd --host=unix:///var/run/docker.sock --host=tcp://127.0.0.1:2375 --storage-driver=overlay2&
+ - timeout 15 sh -c "until docker info; do echo .; sleep 1; done"
+ pre_build:
+ commands:
+ # Add AWSCLI and bash
+ - (apk add --update python python-dev py-pip build-base bash && pip install awscli --upgrade)
+ build:
+ commands:
+ # Build new integration test container
+ - (IMG=$INTEGRATION_CONTAINER_REPOSITORY bash codebuild/scripts/build_deploy_integration_container.sh)

codebuild/scripts/build_deploy_integration_container.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+
+# This script will build the integration test container. This container contains
+# all the tools necessary for running the build and test steps for each of the
+# CodeBuild projects. The script will also tag the container with the latest
+# commit SHA, and with the "latest" tag, then push to an ECR repository.
+
+set -x
+
+# Build new integration test container
+pushd tests
+IMG=$INTEGRATION_CONTAINER_REPOSITORY bash build_integration.sh
+popd
+
+# Log into ECR
+$(aws ecr get-login --no-include-email --region $REGION --registry-ids $AWS_ACCOUNT_ID)
+
+# Tag the container with SHA and latest
+docker tag $INTEGRATION_CONTAINER_REPOSITORY $AWS_ACCOUNT_ID.dkr.ecr.$REGION.amazonaws.com/$INTEGRATION_CONTAINER_REPOSITORY:$CODEBUILD_RESOLVED_SOURCE_VERSION
+docker tag $INTEGRATION_CONTAINER_REPOSITORY $AWS_ACCOUNT_ID.dkr.ecr.$REGION.amazonaws.com/$INTEGRATION_CONTAINER_REPOSITORY:latest
+
+# Push the newly tagged containers
+docker push $AWS_ACCOUNT_ID.dkr.ecr.$REGION.amazonaws.com/$INTEGRATION_CONTAINER_REPOSITORY:$CODEBUILD_RESOLVED_SOURCE_VERSION
+docker push $AWS_ACCOUNT_ID.dkr.ecr.$REGION.amazonaws.com/$INTEGRATION_CONTAINER_REPOSITORY:latest

codebuild/scripts/deployment_constants.sh

@@ -0,0 +1,12 @@
+RELEASE_BUCKET_NAME_FMT="%s-%s"
+
+RELEASE_BINARY_PREFIX_FMT="s3://%s/kubectl-smlogs-plugin"
+ALPHA_BINARY_PREFIX_FMT="s3://%s/%s"
+
+ALPHA_LINUX_BINARY_PATH_FMT="%s/kubectl-smlogs-plugin.linux.amd64.tar.gz"
+ALPHA_DARWIN_BINARY_PATH_FMT="%s/kubectl-smlogs-plugin.darwin.amd64.tar.gz"
+
+RELEASE_LINUX_BINARY_PATH_FMT="%s/%s/linux.amd64.tar.gz"
+RELEASE_DARWIN_BINARY_PATH_FMT="%s/%s/darwin.amd64.tar.gz"
+
+PUBLIC_CP_ARGS="--acl public-read"

codebuild/scripts/package_alpha_operators.sh

@@ -1,9 +1,9 @@
 #!/bin/bash
 
-set -x
-
 source codebuild/scripts/package_operators.sh
 
+set -x
+
 # Login to alpha ECR
 $(aws ecr get-login --no-include-email --region $ALPHA_REPOSITORY_REGION --registry-ids $ALPHA_ACCOUNT_ID)
 

codebuild/scripts/package_operators.sh

@@ -1,7 +1,15 @@
 #!/bin/bash
 
+source codebuild/scripts/deployment_constants.sh
+
 set -e
 
+# Define alpha artifact locations
+printf -v ALPHA_BUCKET_PREFIX $ALPHA_BINARY_PREFIX_FMT $ALPHA_TARBALL_BUCKET $CODEBUILD_RESOLVED_SOURCE_VERSION
+
+printf -v ALPHA_LINUX_BINARY_PATH $ALPHA_LINUX_BINARY_PATH_FMT $ALPHA_BUCKET_PREFIX
+printf -v ALPHA_DARWIN_BINARY_PATH $ALPHA_DARWIN_BINARY_PATH_FMT $ALPHA_BUCKET_PREFIX
+
 # This function deploys a region-specific operator to an ECR prod repo from the existing
 # image in the alpha repository. The function also copies across the smlogs binaries
 # from the alpha tarball bucket into the production buckets.
@@ -34,18 +42,15 @@ function deploy_from_alpha()
  docker push ${dest_ecr_image}:$CODEBUILD_RESOLVED_SOURCE_VERSION
  docker push ${dest_ecr_image}:latest
 
- local bucket_name="${RELEASE_TARBALL_BUCKET_PREFIX}-${account_region}"
- local binary_prefix="s3://${bucket_name}/kubectl-smlogs-plugin"
- local alpha_prefix="s3://$ALPHA_TARBALL_BUCKET/${CODEBUILD_RESOLVED_SOURCE_VERSION}"
-
- local cp_args="--acl public-read"
+ printf -v bucket_name $RELEASE_BUCKET_NAME_FMT $RELEASE_TARBALL_BUCKET_PREFIX $account_region
+ printf -v binary_prefix $RELEASE_BINARY_PREFIX_FMT $bucket_name
 
  # Copy across the binaries and set as latest
- aws s3 cp "${alpha_prefix}/kubectl-smlogs-plugin.linux.amd64.tar.gz" "${binary_prefix}/${CODEBUILD_RESOLVED_SOURCE_VERSION}/linux.amd64.tar.gz" ${cp_args}
- aws s3 cp "${alpha_prefix}/kubectl-smlogs-plugin.linux.amd64.tar.gz" "${binary_prefix}/latest/linux.amd64.tar.gz" ${cp_args}
+ aws s3 cp "$ALPHA_LINUX_BINARY_PATH" "$(printf $RELEASE_LINUX_BINARY_PATH_FMT $binary_prefix $CODEBUILD_RESOLVED_SOURCE_VERSION)" $PUBLIC_CP_ARGS
+ aws s3 cp "$ALPHA_LINUX_BINARY_PATH" "$(printf $RELEASE_LINUX_BINARY_PATH_FMT $binary_prefix latest)" $PUBLIC_CP_ARGS
 
- aws s3 cp "${alpha_prefix}/kubectl-smlogs-plugin.darwin.amd64.tar.gz" "${binary_prefix}/${CODEBUILD_RESOLVED_SOURCE_VERSION}/darwin.amd64.tar.gz" ${cp_args}
- aws s3 cp "${alpha_prefix}/kubectl-smlogs-plugin.darwin.amd64.tar.gz" "${binary_prefix}/latest/darwin.amd64.tar.gz" ${cp_args}
+ aws s3 cp "$ALPHA_DARWIN_BINARY_PATH" "$(printf $RELEASE_DARWIN_BINARY_PATH_FMT $binary_prefix $CODEBUILD_RESOLVED_SOURCE_VERSION)" $PUBLIC_CP_ARGS
+ aws s3 cp "$ALPHA_DARWIN_BINARY_PATH" "$(printf $RELEASE_DARWIN_BINARY_PATH_FMT $binary_prefix latest)" $PUBLIC_CP_ARGS
 }
 
 # This function builds, packages and deploys a region-specific operator to an ECR repo and output bucket.
@@ -109,8 +114,8 @@ function package_operator()
  tar cvzf kubectl-smlogs-plugin.linux.amd64.tar.gz kubectl-smlogs.linux.amd64
  tar cvzf kubectl-smlogs-plugin.darwin.amd64.tar.gz kubectl-smlogs.darwin.amd64
 
- aws s3 cp kubectl-smlogs-plugin.linux.amd64.tar.gz "s3://$ALPHA_TARBALL_BUCKET/${CODEBUILD_RESOLVED_SOURCE_VERSION}/kubectl-smlogs-plugin.linux.amd64.tar.gz"
- aws s3 cp kubectl-smlogs-plugin.darwin.amd64.tar.gz "s3://$ALPHA_TARBALL_BUCKET/${CODEBUILD_RESOLVED_SOURCE_VERSION}/kubectl-smlogs-plugin.darwin.amd64.tar.gz"
+ aws s3 cp kubectl-smlogs-plugin.linux.amd64.tar.gz "$ALPHA_LINUX_BINARY_PATH"
+ aws s3 cp kubectl-smlogs-plugin.darwin.amd64.tar.gz "$ALPHA_DARWIN_BINARY_PATH"
  popd
  fi
 
@@ -119,6 +124,6 @@ function package_operator()
  tar cvzf sagemaker-k8s-operator.tar.gz sagemaker-k8s-operator
 
  # Upload the final tar ball to s3 with standard name and git SHA
- aws s3 cp sagemaker-k8s-operator.tar.gz "s3://$ALPHA_TARBALL_BUCKET/${CODEBUILD_RESOLVED_SOURCE_VERSION}/sagemaker-k8s-operator-${account_region}${tarball_suffix}.tar.gz"
+ aws s3 cp sagemaker-k8s-operator.tar.gz "$ALPHA_BUCKET_PREFIX/sagemaker-k8s-operator-${account_region}${tarball_suffix}.tar.gz"
  popd
 }

codebuild/scripts/release_tag.sh

@@ -1,7 +1,15 @@
 #!/bin/bash
 
+source codebuild/scripts/deployment_constants.sh
+
 set -e
 
+# Define alpha artifact locations
+printf -v ALPHA_BUCKET_PREFIX $ALPHA_BINARY_PREFIX_FMT $ALPHA_TARBALL_BUCKET $CODEBUILD_RESOLVED_SOURCE_VERSION
+
+printf -v ALPHA_LINUX_BINARY_PATH $ALPHA_LINUX_BINARY_PATH_FMT $ALPHA_BUCKET_PREFIX
+printf -v ALPHA_DARWIN_BINARY_PATH $ALPHA_DARWIN_BINARY_PATH_FMT $ALPHA_BUCKET_PREFIX
+
 # This function will pull an existing image + tag and push it with a new tag.
 # Parameter:
 # $1: The repository and image to pull from.
@@ -18,10 +26,26 @@ function retag_image()
  docker push $image:$new_tag
 }
 
-CODEBUILD_GIT_TAG="$(git describe --tags --exact-match 2>/dev/null)"
+# This function will push artifacts to their own folder with a given tag.
+# Parameter:
+# $1: The new tag to push for the artifacts.
+# $2: The region of the new artifacts.
+function retag_binaries()
+{
+ local new_tag="$1"
+ local region="$2"
+
+ printf -v release_bucket $RELEASE_BUCKET_NAME_FMT $RELEASE_TARBALL_BUCKET_PREFIX $region
+ printf -v binary_prefix $RELEASE_BINARY_PREFIX_FMT $release_bucket
+
+ aws s3 cp "$ALPHA_LINUX_BINARY_PATH" "$(printf $RELEASE_LINUX_BINARY_PATH_FMT $binary_prefix $new_tag)" $PUBLIC_CP_ARGS
+ aws s3 cp "$ALPHA_DARWIN_BINARY_PATH" "$(printf $RELEASE_DARWIN_BINARY_PATH_FMT $binary_prefix $new_tag)" $PUBLIC_CP_ARGS
+}
+
+GIT_TAG="$(git describe --tags --exact-match 2>/dev/null)"
 
 # Only run the release process for tagged commits
-if [ "$CODEBUILD_GIT_TAG" == "" ]; then
+if [ "$GIT_TAG" == "" ]; then
  exit 0
 fi
 
@@ -45,9 +69,21 @@ for row in $(echo ${ACCOUNTS_ESCAPED} | jq -r '.[] | @base64'); do
 
  image=${repository_account}.dkr.ecr.${region}.amazonaws.com/${image_repository}
  old_tag="${CODEBUILD_RESOLVED_SOURCE_VERSION}"
- new_tag="${CODEBUILD_GIT_TAG}"
+ full_tag="${GIT_TAG}"
+
+ # Get minor and major version tags
+ [[ $GIT_TAG =~ ^v[0-9]+\.[0-9]+ ]] && minor_tag="${BASH_REMATCH[0]}"
+ [[ $GIT_TAG =~ ^v[0-9]+ ]] && major_tag="${BASH_REMATCH[0]}"
+
+ echo "Tagging $region with $full_tag"
+
+ retag_image "$image" "$old_tag" "$full_tag"
+ retag_image "$image" "$old_tag" "$minor_tag"
+ retag_image "$image" "$old_tag" "$major_tag"
 
- echo "Tagging $image:$old_tag to $image:$new_tag"
+ retag_binaries "$full_tag" "$region"
+ retag_binaries "$minor_tag" "$region"
+ retag_binaries "$major_tag" "$region"
 
- retag_image "$image" "$old_tag" "$new_tag"
+ echo "Finished tagging $region with $full_tag"
 done

controllers/batchtransformjob/batchtransformjob_controller.go

@@ -155,7 +155,7 @@ func (r *BatchTransformJobReconciler) reconcileJob(ctx reconcileRequestContext)
 } else {
 
 ctx.Log.Info("Error getting batchtransformjob state in SageMaker", "requestErr", requestErr)
-return r.handleSageMakerApiFailure(ctx, requestErr)
+return r.handleSageMakerApiFailure(ctx, requestErr, false)
 }
 
 }
@@ -180,12 +180,9 @@ func (r *BatchTransformJobReconciler) reconcileJobDeletion(ctx reconcileRequestC
 } else {
 // Case 2
 log.Info("Sagemaker returns 4xx or 5xx or unrecoverable API Error")
-if requestErr.StatusCode() == 400 {
-// handleSageMakerAPIFailure does not removes the finalizer
-r.removeFinalizerAndUpdate(ctx)
-}
+
 // Handle the 500 or unrecoverable API Error
-return r.handleSageMakerApiFailure(ctx, requestErr)
+return r.handleSageMakerApiFailure(ctx, requestErr, true)
 }
 } else {
 log.Info("Job exists in Sagemaker, lets delete it")
@@ -227,7 +224,7 @@ func (r *BatchTransformJobReconciler) deleteBatchTransformJobIfFinalizerExists(c
 _, err := req.Send(ctx)
 if err != nil {
 log.Error(err, "Unable to stop the job in sagemaker", "context", ctx)
-return r.handleSageMakerApiFailure(ctx, err)
+return r.handleSageMakerApiFailure(ctx, err, false)
 }
 
 return RequeueImmediately()
@@ -301,7 +298,7 @@ func (r *BatchTransformJobReconciler) reconcileSpecWithDescription(ctx reconcile
 return NoRequeue()
 }
 
-func (r *BatchTransformJobReconciler) handleSageMakerApiFailure(ctx reconcileRequestContext, apiErr error) (ctrl.Result, error) {
+func (r *BatchTransformJobReconciler) handleSageMakerApiFailure(ctx reconcileRequestContext, apiErr error, allowRemoveFinalizer bool) (ctrl.Result, error) {
 if err := r.updateJobStatus(ctx, batchtransformjobv1.BatchTransformJobStatus{
 Additional: apiErr.Error(),
 LastCheckTime: Now(),
@@ -316,6 +313,11 @@ func (r *BatchTransformJobReconciler) handleSageMakerApiFailure(ctx reconcileReq
 ctx.Log.Info("SageMaker rate limit exceeded, will retry", "err", awsErr)
 return RequeueAfterInterval(r.PollInterval, nil)
 } else if awsErr.StatusCode() == 400 {
+
+if allowRemoveFinalizer {
+return r.removeFinalizerAndUpdate(ctx)
+}
+
 return NoRequeue()
 } else {
 return RequeueAfterInterval(r.PollInterval, nil)
@@ -357,7 +359,7 @@ func (r *BatchTransformJobReconciler) createBatchTransformJob(ctx reconcileReque
 return RequeueImmediately()
 }
 ctx.Log.Info("Unable to create Transform job", "createError", createError)
-return r.handleSageMakerApiFailure(ctx, createError)
+return r.handleSageMakerApiFailure(ctx, createError, false)
 }
 
 func (r *BatchTransformJobReconciler) getSageMakerDescription(ctx reconcileRequestContext) (*sagemaker.DescribeTransformJobOutput, awserr.RequestFailure) {

controllers/batchtransformjob/batchtransformjob_controller_test.go

@@ -444,6 +444,34 @@ var _ = Describe("Reconciling a job with finalizer that is being deleted", func(
 Expect(job.Status.TransformJobStatus).To(ContainSubstring(string(sagemaker.TransformJobStatusStopping)))
 })
 
+It("should update the status and retry if SageMaker throttles", func() {
+rateExceededMessage := "Rate exceeded"
+// Setup mock responses.
+sageMakerClient := builder.
+AddDescribeTransformJobErrorResponse("ThrottlingException", 400, "request id", rateExceededMessage).
+Build()
+
+// Instantiate controller and reconciliation request.
+controller := createTransformJobReconcilerForSageMakerClient(k8sClient, sageMakerClient, 1)
+request := CreateReconciliationRequest(job.ObjectMeta.Name, job.ObjectMeta.Namespace)
+
+// Run test and verify expectations.
+reconciliationResult, err := controller.Reconcile(request)
+
+Expect(receivedRequests.Len()).To(Equal(1))
+Expect(err).ToNot(HaveOccurred())
+Expect(reconciliationResult.Requeue).To(Equal(false))
+Expect(reconciliationResult.RequeueAfter).To(Equal(controller.PollInterval))
+
+// Verify status is updated.
+err = k8sClient.Get(context.Background(), types.NamespacedName{
+Namespace: job.ObjectMeta.Namespace,
+Name: job.ObjectMeta.Name,
+}, job)
+
+Expect(job.Status.Additional).To(ContainSubstring(rateExceededMessage))
+})
+
 It("should remove the finalizer and not requeue if the job is stopped", func() {
 description.TransformJobStatus = sagemaker.TransformJobStatusStopped
 // Setup mock responses.

controllers/controllertest/mock_sagemaker_client.go

@@ -274,9 +274,9 @@ func (m *MockSageMakerClientBuilder) AddDescribeEndpointErrorResponse(code strin
 }
 
 // Add a DescribeTrainingJob error response to the client.
-func (m *MockSageMakerClientBuilder) AddDescribeTransformJobErrorResponse(code string, statusCode int, reqId string) *MockSageMakerClientBuilder {
+func (m *MockSageMakerClientBuilder) AddDescribeTransformJobErrorResponse(code string, statusCode int, reqId, message string) *MockSageMakerClientBuilder {
 m.responses.PushBack(describeTransformJobResponse{
-err: awserr.NewRequestFailure(awserr.New(code, "mock error message", fmt.Errorf(code)), statusCode, reqId),
+err: awserr.NewRequestFailure(awserr.New(code, message, fmt.Errorf(code)), statusCode, reqId),
 data: nil,
 })
 return m