Replace grpc_health_probe with the built-in gRPC container probe feature

Signed-off-by: Yuki Iwai <[email protected]>
kubeflow · Aug 2, 2023 · db8f9ed · db8f9ed
1 parent e69235d
commit db8f9ed
Show file tree

Hide file tree

Showing 13 changed files with 18 additions and 108 deletions.
diff --git a/cmd/db-manager/v1beta1/Dockerfile b/cmd/db-manager/v1beta1/Dockerfile
@@ -2,7 +2,6 @@
 FROM golang:alpine AS build-env
 
 ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
 
 WORKDIR /go/src/github.com/kubeflow/katib
 
@@ -18,13 +17,8 @@ COPY pkg/ pkg/
 # Build the binary.
 RUN CGO_ENABLED=0 GOOS=linux GOARCH="${TARGETARCH}" go build -a -o katib-db-manager ./cmd/db-manager/v1beta1
 
-# Add GRPC health probe.
-RUN wget -qO /bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 # Copy the db-manager into a thin image.
 FROM alpine:3.15
 WORKDIR /app
-COPY --from=build-env /bin/grpc_health_probe /bin/
 COPY --from=build-env /go/src/github.com/kubeflow/katib/katib-db-manager /app/
 ENTRYPOINT ["./katib-db-manager"]
diff --git a/cmd/suggestion/goptuna/v1beta1/Dockerfile b/cmd/suggestion/goptuna/v1beta1/Dockerfile
@@ -2,7 +2,6 @@
 FROM golang:alpine AS build-env
 
 ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
 
 WORKDIR /go/src/github.com/kubeflow/katib
 
@@ -18,18 +17,13 @@ COPY pkg/ pkg/
 # Build the binary.
 RUN CGO_ENABLED=0 GOOS=linux GOARCH=${TARGETARCH} go build -a -o goptuna-suggestion ./cmd/suggestion/goptuna/v1beta1
 
-# Add GRPC health probe.
-RUN wget -qO/bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 # Copy the Goptuna suggestion into a thin image.
 FROM alpine:3.15
 
 ENV TARGET_DIR /opt/katib
 
 WORKDIR ${TARGET_DIR}
 
-COPY --from=build-env /bin/grpc_health_probe /bin/
 COPY --from=build-env /go/src/github.com/kubeflow/katib/goptuna-suggestion ${TARGET_DIR}/
 
 RUN chgrp -R 0 ${TARGET_DIR} \

diff --git a/cmd/suggestion/hyperband/v1beta1/Dockerfile b/cmd/suggestion/hyperband/v1beta1/Dockerfile
@@ -1,11 +1,3 @@
-FROM alpine:3.15 AS downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO/bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
@@ -22,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ] || [ "${TARGETARCH}" = "arm64" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/cmd/suggestion/hyperopt/v1beta1/Dockerfile b/cmd/suggestion/hyperopt/v1beta1/Dockerfile
@@ -1,11 +1,3 @@
-FROM alpine:3.15 AS downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO/bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
@@ -22,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/cmd/suggestion/nas/darts/v1beta1/Dockerfile b/cmd/suggestion/nas/darts/v1beta1/Dockerfile
@@ -1,11 +1,3 @@
-FROM alpine:3.15 as downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO/bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
@@ -22,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/cmd/suggestion/nas/enas/v1beta1/Dockerfile b/cmd/suggestion/nas/enas/v1beta1/Dockerfile
@@ -1,17 +1,8 @@
-FROM alpine:3.15 AS downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO/bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
 ENV TARGET_DIR /opt/katib
 ENV SUGGESTION_DIR cmd/suggestion/nas/enas/v1beta1
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
 ENV PYTHONPATH ${TARGET_DIR}:${TARGET_DIR}/pkg/apis/manager/v1beta1/python:${TARGET_DIR}/pkg/apis/manager/health/python
 
 RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
@@ -23,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/cmd/suggestion/optuna/v1beta1/Dockerfile b/cmd/suggestion/optuna/v1beta1/Dockerfile
@@ -1,11 +1,3 @@
-FROM alpine:3.15 AS downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO /bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
@@ -22,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/cmd/suggestion/pbt/v1beta1/Dockerfile b/cmd/suggestion/pbt/v1beta1/Dockerfile
@@ -1,11 +1,3 @@
-FROM alpine:3.15 AS downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO /bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
@@ -22,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/cmd/suggestion/skopt/v1beta1/Dockerfile b/cmd/suggestion/skopt/v1beta1/Dockerfile
@@ -1,11 +1,3 @@
-FROM alpine:3.15 AS downloader
-
-ARG TARGETARCH
-ENV GRPC_HEALTH_PROBE_VERSION v0.4.15
-
-RUN wget -qO /bin/grpc_health_probe https://github.com/grpc-ecosystem/grpc-health-probe/releases/download/${GRPC_HEALTH_PROBE_VERSION}/grpc_health_probe-linux-${TARGETARCH} \
-  && chmod +x /bin/grpc_health_probe
-
 FROM python:3.10-slim
 
 ARG TARGETARCH
@@ -22,7 +14,6 @@ RUN if [ "${TARGETARCH}" = "ppc64le" ]; then \
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
-COPY --from=downloader /bin/grpc_health_probe /bin/grpc_health_probe
 
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
 

diff --git a/manifests/v1beta1/components/db-manager/db-manager.yaml b/manifests/v1beta1/components/db-manager/db-manager.yaml
@@ -35,8 +35,8 @@ spec:
             - name: api
               containerPort: 6789
           livenessProbe:
-            exec:
-              command: ["/bin/grpc_health_probe", "-addr=:6789"]
+            grpc:
+              port: 6789
             initialDelaySeconds: 10
             periodSeconds: 60
             failureThreshold: 5
diff --git a/pkg/controller.v1beta1/consts/const.go b/pkg/controller.v1beta1/consts/const.go
@@ -87,10 +87,6 @@ const (
 	// DefaultEarlyStoppingPort is the default port of EarlyStopping service.
 	DefaultEarlyStoppingPort = 6788
 
-	// DefaultGRPCService is the default suggestion service name,
-	// which is used to run healthz check using grpc probe.
-	DefaultGRPCService = "manager.v1beta1.Suggestion"
-
 	// DefaultGRPCRetryAttempts is the the maximum number of retries for gRPC calls
 	DefaultGRPCRetryAttempts = 10
 	// DefaultGRPCRetryPeriod is a fixed period of time between gRPC call retries
@@ -171,6 +167,10 @@ var (
 	// DefaultKatibDBManagerServicePort is the default Port of Katib DB Manager
 	DefaultKatibDBManagerServicePort = env.GetEnvOrDefault(DefaultKatibDBManagerServicePortEnvName, "6789")
 
+	// DefaultGRPCService is the default suggestion service name,
+	// which is used to run healthz check using grpc probe.
+	DefaultGRPCService = "manager.v1beta1.Suggestion"
+
 	// List of all valid keys of trial metadata for substitution in Trial template
 	TrialTemplateMetaKeys = []string{
 		TrialTemplateMetaKeyOfName,

diff --git a/pkg/controller.v1beta1/suggestion/composer/composer.go b/pkg/controller.v1beta1/suggestion/composer/composer.go
@@ -44,8 +44,6 @@ const (
 	defaultPeriodForReady      = 10
 	defaultPeriodForLive       = 120
 	defaultFailureThreshold    = 12
-	// Ref https://github.com/grpc-ecosystem/grpc-health-probe/
-	defaultGRPCHealthCheckProbe = "/bin/grpc_health_probe"
 )
 
 var (
@@ -210,12 +208,9 @@ func (g *General) desiredContainers(s *suggestionsv1beta1.Suggestion,
 	if viper.GetBool(consts.ConfigEnableGRPCProbeInSuggestion) && suggestionContainer.ReadinessProbe == nil {
 		suggestionContainer.ReadinessProbe = &corev1.Probe{
 			ProbeHandler: corev1.ProbeHandler{
-				Exec: &corev1.ExecAction{
-					Command: []string{
-						defaultGRPCHealthCheckProbe,
-						fmt.Sprintf("-addr=:%d", consts.DefaultSuggestionPort),
-						fmt.Sprintf("-service=%s", consts.DefaultGRPCService),
-					},
+				GRPC: &corev1.GRPCAction{
+					Port:    consts.DefaultSuggestionPort,
+					Service: &consts.DefaultGRPCService,
 				},
 			},
 			InitialDelaySeconds: defaultInitialDelaySeconds,
@@ -225,12 +220,9 @@ func (g *General) desiredContainers(s *suggestionsv1beta1.Suggestion,
 	if viper.GetBool(consts.ConfigEnableGRPCProbeInSuggestion) && suggestionContainer.LivenessProbe == nil {
 		suggestionContainer.LivenessProbe = &corev1.Probe{
 			ProbeHandler: corev1.ProbeHandler{
-				Exec: &corev1.ExecAction{
-					Command: []string{
-						defaultGRPCHealthCheckProbe,
-						fmt.Sprintf("-addr=:%d", consts.DefaultSuggestionPort),
-						fmt.Sprintf("-service=%s", consts.DefaultGRPCService),
-					},
+				GRPC: &corev1.GRPCAction{
+					Port:    consts.DefaultSuggestionPort,
+					Service: &consts.DefaultGRPCService,
 				},
 			},
 			// Ref https://srcco.de/posts/kubernetes-liveness-probes-are-dangerous.html

diff --git a/pkg/controller.v1beta1/suggestion/composer/composer_test.go b/pkg/controller.v1beta1/suggestion/composer/composer_test.go
@@ -817,25 +817,19 @@ func newFakeContainers() []corev1.Container {
 			},
 			ReadinessProbe: &corev1.Probe{
 				ProbeHandler: corev1.ProbeHandler{
-					Exec: &corev1.ExecAction{
-						Command: []string{
-							defaultGRPCHealthCheckProbe,
-							fmt.Sprintf("-addr=:%d", consts.DefaultSuggestionPort),
-							fmt.Sprintf("-service=%s", consts.DefaultGRPCService),
-						},
+					GRPC: &corev1.GRPCAction{
+						Port:    consts.DefaultSuggestionPort,
+						Service: &consts.DefaultGRPCService,
 					},
 				},
 				InitialDelaySeconds: defaultInitialDelaySeconds,
 				PeriodSeconds:       defaultPeriodForReady,
 			},
 			LivenessProbe: &corev1.Probe{
 				ProbeHandler: corev1.ProbeHandler{
-					Exec: &corev1.ExecAction{
-						Command: []string{
-							defaultGRPCHealthCheckProbe,
-							fmt.Sprintf("-addr=:%d", consts.DefaultSuggestionPort),
-							fmt.Sprintf("-service=%s", consts.DefaultGRPCService),
-						},
+					GRPC: &corev1.GRPCAction{
+						Port:    consts.DefaultSuggestionPort,
+						Service: &consts.DefaultGRPCService,
 					},
 				},
 				InitialDelaySeconds: defaultInitialDelaySeconds,