*: fix "eks/amazon-eks-ami-issue-454", release (#197)

* *: fix "eks/amazon-eks-ami-issue-454"

Signed-off-by: Gyuho Lee <[email protected]>

* CHANGELOG: update

Signed-off-by: Gyuho Lee <[email protected]>

* hack/build.sh: add "arm64" builds

Signed-off-by: Gyuho Lee <[email protected]>

CHANGELOG/CHANGELOG-1.5.md

@@ -3,10 +3,14 @@
 
 <hr>
 
-## [v1.5.6](https://github.com/aws/aws-k8s-tester/releases/tag/v1.5.6) (2021-01)
+## [v1.5.6](https://github.com/aws/aws-k8s-tester/releases/tag/v1.5.6) (2021-01-25)
 
 See [code changes](https://github.com/aws/aws-k8s-tester/compare/v1.5.5...v1.5.6).
 
+### Release
+
+- Add [arm64 (Apple M1) builds](https://github.com/aws/aws-k8s-tester/pull/193).
+
 ### `eksconfig`
 
 - Expose [configuration of sonobuoy worker/systemd-logs container](https://github.com/aws/aws-k8s-tester/pull/190).
@@ -16,14 +20,16 @@ See [code changes](https://github.com/aws/aws-k8s-tester/compare/v1.5.5...v1.5.6
 
 - Fix [`eks/mng` delete retries](https://github.com/aws/aws-k8s-tester/pull/196).
   - See https://github.com/aws/aws-k8s-tester/issues/195.
+- Add [`eks/amazon-eks-ami-issue-454`](https://github.com/aws/aws-k8s-tester/pull/193).
+  - Fix [list nodes](https://github.com/aws/aws-k8s-tester/pull/197).
 
 ### Dependency
 
 - Upgrade [`github.com/aws/aws-sdk-go`](https://github.com/aws/aws-sdk-go/releases) from [`v1.35.27`](https://github.com/aws/aws-sdk-go/releases/tag/v1.35.27) to [`v1.35.30`](https://github.com/aws/aws-sdk-go/releases/tag/v1.35.30).
   - `v1.36.2` breaks darwin builds.
 ### Go
 
-- Compile with [*Go 1.15.6*](https://golang.org/doc/devel/release.html#go1.15).
+- Compile with [*Go 1.16beta1*](https://golang.org/doc/devel/release.html#go1.16).
 
 
 

eks/amazon-eks-ami-issue-454/soft-lockup-454.go

@@ -275,10 +275,13 @@ func (ts *tester) waitDeployment() (err error) {
 
 func (ts *tester) validateNodesStayHealthy() (err error) {
 	nodeSelector := ts.getNodeSelector()
-
-	nodes, err := ts.cfg.K8SClient.ListNodesWithSelector(1000, 5*time.Second, nodeSelector)
+	nodes, err := ts.cfg.K8SClient.ListNodes(
+		1000,
+		5*time.Second,
+		k8s_client.WithFieldSelector(nodeSelector),
+	)
 	if err != nil {
-		ts.cfg.Logger.Warn("get nodes failed", zap.Error(err))
+		ts.cfg.Logger.Warn("list nodes failed", zap.Error(err))
 		return err
 	}
 

hack/build.sh

@@ -13,32 +13,40 @@ echo "GIT_COMMIT:" ${GIT_COMMIT}
 echo "RELEASE_VERSION:" ${RELEASE_VERSION}
 echo "BUILD_TIME:" ${BUILD_TIME}
 
-DEFAULT_WHAT='aws-k8s-tester cw-utils ec2-utils ecr-utils eks-utils etcd-utils s3-utils sts-utils'
+DEFAULT_ARCHS='amd64 arm64'
 DEFAULT_TARGETS='linux darwin'
+DEFAULT_WHAT='aws-k8s-tester cw-utils ec2-utils ecr-utils eks-utils etcd-utils s3-utils sts-utils'
 
-PACKAGE_NAME='github.com/aws/aws-k8s-tester'
-WHAT=${WHAT:-$DEFAULT_WHAT}
+ARCHS=${ARCHS:-$DEFAULT_ARCHS}
 TARGETS=${TARGETS:-$DEFAULT_TARGETS}
+WHAT=${WHAT:-$DEFAULT_WHAT}
 
 echo ""
 echo "Usage: \`make TARGETS='linux' WHAT='aws-k8s-tester cw-utils'\`"
+echo "DEFAULT_ARCHS=$DEFAULT_ARCHS"
 echo "DEFAULT_TARGETS=$DEFAULT_TARGETS"
 echo "DEFAULT_WHAT=$DEFAULT_WHAT"
 echo ""
 
 mkdir -p ./bin
 
-for os in ${TARGETS}; do
-  for cmd in ${WHAT}; do
-    echo "=== Building target=${cmd}, os=${os} ==="
-    CGO_ENABLED=0 GOOS=${os} GOARCH=$(go env GOARCH) \
-      go build -mod=vendor -v \
-      -ldflags "-s -w \
-      -X ${PACKAGE_NAME}/version.GitCommit=${GIT_COMMIT} \
-      -X ${PACKAGE_NAME}/version.ReleaseVersion=${RELEASE_VERSION} \
-      -X ${PACKAGE_NAME}/version.BuildTime=${BUILD_TIME}" \
-      -o ./bin/${cmd}-${RELEASE_VERSION}-${os}-$(go env GOARCH) \
-      ./cmd/${cmd}
+PACKAGE_NAME='github.com/aws/aws-k8s-tester'
+for arch in ${ARCHS}; do
+  for os in ${TARGETS}; do
+    for bin in ${WHAT}; do
+      echo "=== Building arch=${arch}, os=${os}, target=${bin} ==="
+      CGO_ENABLED=0 \
+        GOARCH=${arch} \
+        GOOS=${os} \
+        go build \
+        -mod=vendor -v \
+        -ldflags "-s -w \
+        -X ${PACKAGE_NAME}/version.GitCommit=${GIT_COMMIT} \
+        -X ${PACKAGE_NAME}/version.ReleaseVersion=${RELEASE_VERSION} \
+        -X ${PACKAGE_NAME}/version.BuildTime=${BUILD_TIME}" \
+        -o ./bin/${bin}-${RELEASE_VERSION}-${os}-$(go env GOARCH) \
+        ./cmd/${bin}
+    done
   done
 done
 

pkg/k8s-client/eks.go

@@ -164,11 +164,11 @@ type EKS interface {
 	// ListNamespaces returns the list of existing namespace names.
 	ListNamespaces(batchLimit int64, batchInterval time.Duration) ([]v1.Namespace, error)
 	// ListNodes returns the list of existing nodes.
-	ListNodes(batchLimit int64, batchInterval time.Duration) ([]v1.Node, error)
-	// ListNodes returns the list of existing CSRs.
+	ListNodes(batchLimit int64, batchInterval time.Duration, opts ...OpOption) ([]v1.Node, error)
+	// ListCSRs returns the list of existing CSRs.
 	ListCSRs(batchLimit int64, batchInterval time.Duration) ([]certificatesv1beta1.CertificateSigningRequest, error)
 	// ListPods returns the list of existing namespace names.
-	ListPods(namespace string, batchLimit int64, batchInterval time.Duration) ([]v1.Pod, error)
+	ListPods(namespace string, batchLimit int64, batchInterval time.Duration, opts ...OpOption) ([]v1.Pod, error)
 	// ListConfigMaps returns the list of existing config maps.
 	ListConfigMaps(namespace string, batchLimit int64, batchInterval time.Duration) ([]v1.ConfigMap, error)
 	// ListSecrets returns the list of existing Secret objects.
@@ -917,20 +917,31 @@ func (e *eks) listNamespaces(batchLimit int64, batchInterval time.Duration) (ns
 	return ns, err
 }
 
-func (e *eks) ListNodes(batchLimit int64, batchInterval time.Duration) ([]v1.Node, error) {
-	ns, err := e.listNodes(batchLimit, batchInterval)
+func (e *eks) ListNodes(batchLimit int64, batchInterval time.Duration, opts ...OpOption) ([]v1.Node, error) {
+	ns, err := e.listNodes(batchLimit, batchInterval, opts...)
 	return ns, err
 }
 
-func (e *eks) listNodes(batchLimit int64, batchInterval time.Duration) (nodes []v1.Node, err error) {
+func (e *eks) listNodes(batchLimit int64, batchInterval time.Duration, opts ...OpOption) (nodes []v1.Node, err error) {
+	ret := Op{}
+	ret.applyOpts(opts)
+
 	e.cfg.Logger.Info("listing nodes",
 		zap.Int64("batch-limit", batchLimit),
 		zap.Duration("batch-interval", batchInterval),
+		zap.String("label-selector", ret.labelSelector),
+		zap.String("field-selector", ret.fieldSelector),
 	)
 	rs := &v1.NodeList{ListMeta: metav1.ListMeta{Continue: ""}}
 	for {
 		ctx, cancel := context.WithTimeout(context.Background(), time.Minute)
-		rs, err = e.getClient().CoreV1().Nodes().List(ctx, metav1.ListOptions{Limit: batchLimit, Continue: rs.Continue})
+		rs, err = e.getClient().CoreV1().Nodes().List(ctx, metav1.ListOptions{
+			Limit:    batchLimit,
+			Continue: rs.Continue,
+
+			LabelSelector: ret.labelSelector,
+			FieldSelector: ret.fieldSelector,
+		})
 		cancel()
 		if err != nil {
 			return nil, err
@@ -986,20 +997,26 @@ func (e *eks) listCSRs(batchLimit int64, batchInterval time.Duration) (csrs []ce
 	return csrs, err
 }
 
-func (e *eks) ListPods(namespace string, batchLimit int64, batchInterval time.Duration) ([]v1.Pod, error) {
-	ns, err := e.listPods(namespace, batchLimit, batchInterval, 5)
+func (e *eks) ListPods(namespace string, batchLimit int64, batchInterval time.Duration, opts ...OpOption) ([]v1.Pod, error) {
+	ns, err := e.listPods(namespace, batchLimit, batchInterval, 5, opts...)
 	return ns, err
 }
 
 func (e *eks) listPods(
 	namespace string,
 	batchLimit int64,
 	batchInterval time.Duration,
-	retryLeft int) (pods []v1.Pod, err error) {
+	retryLeft int,
+	opts ...OpOption) (pods []v1.Pod, err error) {
+	ret := Op{}
+	ret.applyOpts(opts)
+
 	e.cfg.Logger.Info("listing pods",
 		zap.String("namespace", namespace),
 		zap.Int64("batch-limit", batchLimit),
 		zap.Duration("batch-interval", batchInterval),
+		zap.String("label-selector", ret.labelSelector),
+		zap.String("field-selector", ret.fieldSelector),
 	)
 	rs := &v1.PodList{ListMeta: metav1.ListMeta{Continue: ""}}
 	for {
@@ -1014,7 +1031,7 @@ func (e *eks) listPods(
 				// e.g. The provided continue parameter is too old to display a consistent list result. You can start a new list without the continue parameter, or use the continue token in this response to retrieve the remainder of the results. Continuing with the provided token results in an inconsistent list - objects that were created, modified, or deleted between the time the first chunk was returned and now may show up in the list.
 				e.cfg.Logger.Warn("stale list response, retrying for consistent list", zap.Error(err))
 				time.Sleep(15 * time.Second)
-				return e.listPods(namespace, batchLimit, batchInterval, retryLeft-1)
+				return e.listPods(namespace, batchLimit, batchInterval, retryLeft-1, opts...)
 			}
 			return nil, err
 		}

pkg/k8s-client/k8s.go

@@ -911,15 +911,27 @@ func WaitForDeploymentCompletes(
 
 // Op represents a Kubernetes client operation.
 type Op struct {
+	labelSelector   string
+	fieldSelector   string
 	queryFunc       func()
 	podFunc         func(v1.Pod)
 	forceDelete     bool
 	forceDeleteFunc func()
 }
 
-// OpOption configures archiver operations.
+// OpOption configures Kubernetes client operations.
 type OpOption func(*Op)
 
+// WithLabelSelector configures label selector for list operations.
+func WithLabelSelector(s string) OpOption {
+	return func(op *Op) { op.labelSelector = s }
+}
+
+// WithFieldSelector configures field selector for list operations.
+func WithFieldSelector(s string) OpOption {
+	return func(op *Op) { op.fieldSelector = s }
+}
+
 // WithQueryFunc configures query function to be called in retry func.
 func WithQueryFunc(f func()) OpOption {
 	return func(op *Op) { op.queryFunc = f }