Add cpu limit to resource requests and refactor knative service struct

caraml-dev · May 23, 2024 · 916ed3f · 916ed3f
1 parent 70da2bc
commit 916ed3f
Show file tree

Hide file tree

Showing 11 changed files with 302 additions and 213 deletions.
diff --git a/api/turing/cluster/knative_service.go b/api/turing/cluster/knative_service.go
@@ -50,9 +50,7 @@ type KnativeService struct {
 	TopologySpreadConstraints []corev1.TopologySpreadConstraint `json:"topologySpreadConstraints"`
 
 	// Resource properties
-	QueueProxyResourcePercentage          int     `json:"queueProxyResourcePercentage"`
-	UserContainerCPULimitRequestFactor    float64 `json:"userContainerLimitCPURequestFactor"`
-	UserContainerMemoryLimitRequestFactor float64 `json:"userContainerLimitMemoryRequestFactor"`
+	QueueProxyResourcePercentage int `json:"queueProxyResourcePercentage"`
 }
 
 // Creates a new config object compatible with the knative serving API, from
@@ -131,12 +129,6 @@ func (cfg *KnativeService) buildSvcSpec(
 	// Revision name
 	revisionName := getDefaultRevisionName(cfg.Name)
 
-	// Build resource requirements for the user container
-	resourceReqs := cfg.buildResourceReqs(
-		cfg.UserContainerCPULimitRequestFactor,
-		cfg.UserContainerMemoryLimitRequestFactor,
-	)
-
 	// Build container spec
 	var portName string
 	// If protocol is using GRPC, add "h2c" which is required for grpc knative
@@ -151,7 +143,7 @@ func (cfg *KnativeService) buildSvcSpec(
 				ContainerPort: cfg.ContainerPort,
 			},
 		},
-		Resources:    resourceReqs,
+		Resources:    cfg.buildResourceReqs(),
 		VolumeMounts: cfg.VolumeMounts,
 		Env:          cfg.Envs,
 	}

diff --git a/api/turing/cluster/kubernetes_service.go b/api/turing/cluster/kubernetes_service.go
@@ -73,7 +73,7 @@ func (cfg *KubernetesService) buildStatefulSet(labels map[string]string) *appsv1
 							Args:            cfg.Command,
 							Ports:           cfg.buildContainerPorts(),
 							Env:             cfg.Envs,
-							Resources:       cfg.buildResourceReqs(defaultCPULimitRequestFactor, defaultMemoryLimitRequestFactor),
+							Resources:       cfg.buildResourceReqs(),
 							VolumeMounts:    cfg.VolumeMounts,
 							LivenessProbe:   cfg.buildContainerProbe(livenessProbeType, int(cfg.ProbePort)),
 							ReadinessProbe:  cfg.buildContainerProbe(readinessProbeType, int(cfg.ProbePort)),

diff --git a/api/turing/cluster/models.go b/api/turing/cluster/models.go
@@ -35,8 +35,10 @@ type BaseService struct {
 	Image     string `json:"image"`
 
 	// Resources
-	CPURequests    resource.Quantity `json:"cpu_requests"`
-	MemoryRequests resource.Quantity `json:"memory_requests"`
+	CPURequests    resource.Quantity  `json:"cpu_requests"`
+	CPULimit       *resource.Quantity `json:"cpu_limit"`
+	MemoryRequests resource.Quantity  `json:"memory_requests"`
+	MemoryLimit    *resource.Quantity `json:"memory_limit"`
 
 	// Health Checks
 	ProbePort             int32  `json:"probe_port"`
@@ -62,22 +64,19 @@ type BaseService struct {
 	InitContainers []Container `json:"init_containers"`
 }
 
-func (cfg *BaseService) buildResourceReqs(
-	UserContainerCPULimitRequestFactor float64,
-	UserContainerMemoryLimitRequestFactor float64,
-) corev1.ResourceRequirements {
+func (cfg *BaseService) buildResourceReqs() corev1.ResourceRequirements {
 	reqs := map[corev1.ResourceName]resource.Quantity{
 		corev1.ResourceCPU:    cfg.CPURequests,
 		corev1.ResourceMemory: cfg.MemoryRequests,
 	}
 
 	// Set resource limits to request * userContainerCPULimitRequestFactor or UserContainerMemoryLimitRequestFactor
 	limits := map[corev1.ResourceName]resource.Quantity{}
-	if UserContainerCPULimitRequestFactor != 0 {
-		limits[corev1.ResourceCPU] = ComputeResource(cfg.CPURequests, UserContainerCPULimitRequestFactor)
+	if cfg.CPULimit != nil {
+		limits[corev1.ResourceCPU] = *cfg.CPULimit
 	}
-	if UserContainerMemoryLimitRequestFactor != 0 {
-		limits[corev1.ResourceMemory] = ComputeResource(cfg.MemoryRequests, UserContainerMemoryLimitRequestFactor)
+	if cfg.MemoryLimit != nil {
+		limits[corev1.ResourceMemory] = *cfg.MemoryLimit
 	}
 
 	return corev1.ResourceRequirements{

diff --git a/api/turing/cluster/servicebuilder/router.go b/api/turing/cluster/servicebuilder/router.go
@@ -138,7 +138,9 @@ func (sb *clusterSvcBuilder) NewRouterService(
 			Namespace:            namespace,
 			Image:                routerVersion.Image,
 			CPURequests:          routerVersion.ResourceRequest.CPURequest,
+			CPULimit:             sb.getCPULimit(routerVersion.ResourceRequest),
 			MemoryRequests:       routerVersion.ResourceRequest.MemoryRequest,
+			MemoryLimit:          sb.getMemoryLimit(routerVersion.ResourceRequest),
 			LivenessHTTPGetPath:  routerLivenessPath,
 			ReadinessHTTPGetPath: routerReadinessPath,
 			Envs:                 envs,
@@ -148,18 +150,16 @@ func (sb *clusterSvcBuilder) NewRouterService(
 			VolumeMounts:         volumeMounts,
 			InitContainers:       initContainers,
 		},
-		IsClusterLocal:                        false,
-		ContainerPort:                         routerPort,
-		Protocol:                              routerVersion.Protocol,
-		MinReplicas:                           routerVersion.ResourceRequest.MinReplica,
-		MaxReplicas:                           routerVersion.ResourceRequest.MaxReplica,
-		InitialScale:                          initialScale,
-		AutoscalingMetric:                     string(routerVersion.AutoscalingPolicy.Metric),
-		AutoscalingTarget:                     routerVersion.AutoscalingPolicy.Target,
-		TopologySpreadConstraints:             topologySpreadConstraints,
-		QueueProxyResourcePercentage:          sb.knativeServiceConfig.QueueProxyResourcePercentage,
-		UserContainerCPULimitRequestFactor:    sb.knativeServiceConfig.UserContainerCPULimitRequestFactor,
-		UserContainerMemoryLimitRequestFactor: sb.knativeServiceConfig.UserContainerMemoryLimitRequestFactor,
+		IsClusterLocal:               false,
+		ContainerPort:                routerPort,
+		Protocol:                     routerVersion.Protocol,
+		MinReplicas:                  routerVersion.ResourceRequest.MinReplica,
+		MaxReplicas:                  routerVersion.ResourceRequest.MaxReplica,
+		InitialScale:                 initialScale,
+		AutoscalingMetric:            string(routerVersion.AutoscalingPolicy.Metric),
+		AutoscalingTarget:            routerVersion.AutoscalingPolicy.Target,
+		TopologySpreadConstraints:    topologySpreadConstraints,
+		QueueProxyResourcePercentage: sb.knativeServiceConfig.QueueProxyResourcePercentage,
 	}
 	return sb.validateKnativeService(svc)
 }
@@ -210,17 +210,20 @@ func (sb *clusterSvcBuilder) buildRouterEnvs(
 	sentryDSN string,
 	ver *models.RouterVersion,
 ) ([]corev1.EnvVar, error) {
+	envs := sb.getEnvVars(ver.ResourceRequest, nil)
+
 	// Add app name, router timeout, jaeger collector
-	envs := []corev1.EnvVar{
-		{Name: envAppName, Value: fmt.Sprintf("%s-%d.%s", ver.Router.Name, ver.Version, namespace)},
-		{Name: envAppEnvironment, Value: environmentType},
-		{Name: envRouterTimeout, Value: ver.Timeout},
-		{Name: envJaegerEndpoint, Value: routerDefaults.JaegerCollectorEndpoint},
-		{Name: envRouterConfigFile, Value: routerConfigMapMountPath + routerConfigFileName},
-		{Name: envRouterProtocol, Value: string(ver.Protocol)},
-		{Name: envSentryEnabled, Value: strconv.FormatBool(sentryEnabled)},
-		{Name: envSentryDSN, Value: sentryDSN},
-	}
+	envs = append(envs,
+		[]corev1.EnvVar{
+			{Name: envAppName, Value: fmt.Sprintf("%s-%d.%s", ver.Router.Name, ver.Version, namespace)},
+			{Name: envAppEnvironment, Value: environmentType},
+			{Name: envRouterTimeout, Value: ver.Timeout},
+			{Name: envJaegerEndpoint, Value: routerDefaults.JaegerCollectorEndpoint},
+			{Name: envRouterConfigFile, Value: routerConfigMapMountPath + routerConfigFileName},
+			{Name: envRouterProtocol, Value: string(ver.Protocol)},
+			{Name: envSentryEnabled, Value: strconv.FormatBool(sentryEnabled)},
+			{Name: envSentryDSN, Value: sentryDSN},
+		}...)
 
 	// Add enricher / ensembler related env vars, if enabled
 	if ver.Enricher != nil {