Add new max allowed replicas field to env configs and relevant checks

caraml-dev · Mar 1, 2024 · 07b5021 · 07b5021
1 parent a810ebf
commit 07b5021
Show file tree

Hide file tree

Showing 7 changed files with 55 additions and 10 deletions.
diff --git a/api/cluster/controller.go b/api/cluster/controller.go
@@ -195,6 +195,13 @@ func (c *controller) Deploy(ctx context.Context, modelService *models.Service) (
 			log.Errorf("insufficient available memory resource to fulfil user request of %d", memRequest)
 			return nil, ErrInsufficientMem
 		}
+		if modelService.ResourceRequest.MaxReplica > c.deploymentConfig.MaxAllowedReplica {
+			log.Errorf("Requested Max Replica (%d) is more than max permissible (%d)",
+				modelService.ResourceRequest.MaxReplica,
+				c.deploymentConfig.MaxAllowedReplica,
+			)
+			return nil, ErrRequestedMaxReplicasNotAllowed
+		}
 	}
 
 	_, err := c.namespaceCreator.CreateNamespace(ctx, modelService.Namespace)

diff --git a/api/cluster/controller_test.go b/api/cluster/controller_test.go
@@ -655,6 +655,35 @@ func TestController_DeployInferenceService(t *testing.T) {
 			createVsResult:  &vsReactor{vs, nil},
 			wantError:       true,
 		},
+		{
+			name: "error: deploying service due to max replica requests greater than max value allowed",
+			modelService: &models.Service{
+				Name:      isvcName,
+				Namespace: project.Name,
+				Options:   modelOpt,
+				ResourceRequest: &models.ResourceRequest{
+					MinReplica:    2,
+					MaxReplica:    5,
+					CPURequest:    resource.MustParse("1000m"),
+					MemoryRequest: resource.MustParse("1Gi"),
+				},
+			},
+			createResult: &inferenceServiceReactor{
+				&kservev1beta1.InferenceService{ObjectMeta: metav1.ObjectMeta{Name: isvcName, Namespace: project.Name}},
+				nil,
+			},
+			checkResult: &inferenceServiceReactor{
+				&kservev1beta1.InferenceService{
+					ObjectMeta: metav1.ObjectMeta{Name: isvcName, Namespace: project.Name},
+					Status:     statusReady,
+				},
+				nil,
+			},
+			deployTimeout:   deployTimeout,
+			createPdbResult: &pdbReactor{pdb, nil},
+			createVsResult:  &vsReactor{vs, nil},
+			wantError:       true,
+		},
 	}
 
 	for _, tt := range tests {
@@ -696,6 +725,7 @@ func TestController_DeployInferenceService(t *testing.T) {
 				NamespaceTimeout:                   2 * tickDurationSecond * time.Second,
 				MaxCPU:                             resource.MustParse("8"),
 				MaxMemory:                          resource.MustParse("8Gi"),
+				MaxAllowedReplica:                  4,
 				DefaultModelResourceRequests:       &config.ResourceRequests{},
 				DefaultTransformerResourceRequests: &config.ResourceRequests{},
 				PodDisruptionBudget: config.PodDisruptionBudgetConfig{

diff --git a/api/cluster/errors.go b/api/cluster/errors.go
@@ -19,6 +19,7 @@ import "errors"
 var (
 	ErrInsufficientCPU                        = errors.New("CPU request is too large")
 	ErrInsufficientMem                        = errors.New("memory request too large")
+	ErrRequestedMaxReplicasNotAllowed         = errors.New("requested max replicas is more than max permissible")
 	ErrTimeoutNamespace                       = errors.New("timeout creating namespace")
 	ErrUnableToCreateNamespace                = errors.New("error creating namespace")
 	ErrUnableToGetNamespaceStatus             = errors.New("error retrieving namespace status")

diff --git a/api/cmd/api/setup.go b/api/cmd/api/setup.go
@@ -209,14 +209,15 @@ func initEnvironmentService(cfg *config.Config, db *gorm.DB) service.Environment
 
 			log.Infof("adding environment %s: cluster: %s, is_default: %v", envCfg.Name, envCfg.Cluster, envCfg.IsDefault)
 			env = &models.Environment{
-				Name:       envCfg.Name,
-				Cluster:    envCfg.Cluster,
-				IsDefault:  isDefault,
-				Region:     envCfg.Region,
-				GcpProject: envCfg.GcpProject,
-				MaxCPU:     envCfg.MaxCPU,
-				MaxMemory:  envCfg.MaxMemory,
-				GPUs:       models.ParseGPUsConfig(envCfg.GPUs),
+				Name:              envCfg.Name,
+				Cluster:           envCfg.Cluster,
+				IsDefault:         isDefault,
+				Region:            envCfg.Region,
+				GcpProject:        envCfg.GcpProject,
+				MaxCPU:            envCfg.MaxCPU,
+				MaxMemory:         envCfg.MaxMemory,
+				MaxAllowedReplica: envCfg.MaxAllowedReplica,
+				GPUs:              models.ParseGPUsConfig(envCfg.GPUs),
 				DefaultResourceRequest: &models.ResourceRequest{
 					MinReplica:    deploymentCfg.DefaultModelResourceRequests.MinReplica,
 					MaxReplica:    deploymentCfg.DefaultModelResourceRequests.MaxReplica,
@@ -253,6 +254,7 @@ func initEnvironmentService(cfg *config.Config, db *gorm.DB) service.Environment
 			env.GcpProject = envCfg.GcpProject
 			env.MaxCPU = envCfg.MaxCPU
 			env.MaxMemory = envCfg.MaxMemory
+			env.MaxAllowedReplica = envCfg.MaxAllowedReplica
 			env.GPUs = models.ParseGPUsConfig(envCfg.GPUs)
 			env.DefaultResourceRequest = &models.ResourceRequest{
 				MinReplica:    deploymentCfg.DefaultModelResourceRequests.MinReplica,

diff --git a/api/config/deployment.go b/api/config/deployment.go
@@ -31,10 +31,12 @@ type DeploymentConfig struct {
 	DefaultModelResourceRequests *ResourceRequests
 	// Default resource request for transformer deployment
 	DefaultTransformerResourceRequests *ResourceRequests
-	// Max CPU of machine
+	// Max allowed CPU of each pod
 	MaxCPU resource.Quantity
-	// Max Memory of machine
+	// Max allowed Memory of each pod
 	MaxMemory resource.Quantity
+	// Max allowed MaxReplica value of each deployment
+	MaxAllowedReplica int
 	// TopologySpreadConstraints to be applied on the pods of each model deployment
 	TopologySpreadConstraints []corev1.TopologySpreadConstraint
 	// Percentage of knative's queue proxy resource request from the inference service resource request

diff --git a/api/config/environment.go b/api/config/environment.go
@@ -44,6 +44,7 @@ type EnvironmentConfig struct {
 
 	MaxCPU                    string                    `yaml:"max_cpu"`
 	MaxMemory                 string                    `yaml:"max_memory"`
+	MaxAllowedReplica         int                       `yaml:"max_allowed_replica"`
 	TopologySpreadConstraints TopologySpreadConstraints `yaml:"topology_spread_constraints"`
 	PodDisruptionBudget       PodDisruptionBudgetConfig `yaml:"pod_disruption_budget"`
 
@@ -177,6 +178,7 @@ func ParseDeploymentConfig(envCfg *EnvironmentConfig, cfg *Config) DeploymentCon
 		},
 		MaxCPU:                                resource.MustParse(envCfg.MaxCPU),
 		MaxMemory:                             resource.MustParse(envCfg.MaxMemory),
+		MaxAllowedReplica:                     envCfg.MaxAllowedReplica,
 		TopologySpreadConstraints:             envCfg.TopologySpreadConstraints,
 		QueueResourcePercentage:               envCfg.QueueResourcePercentage,
 		PyfuncGRPCOptions:                     cfg.PyfuncGRPCOptions,

diff --git a/api/models/environment.go b/api/models/environment.go
@@ -23,6 +23,7 @@ type Environment struct {
 	GcpProject                        string           `json:"gcp_project"`
 	MaxCPU                            string           `json:"max_cpu"`
 	MaxMemory                         string           `json:"max_memory"`
+	MaxAllowedReplica                 int              `yaml:"max_allowed_replica"`
 	GPUs                              GPUs             `json:"gpus" gorm:"column:gpus"`
 	DefaultResourceRequest            *ResourceRequest `json:"default_resource_request"`
 	DefaultTransformerResourceRequest *ResourceRequest `json:"default_transformer_resource_request"`