Make instance volume size configurable

Author: deliahu

cli/cmd/lib_cluster_config.go

@@ -219,6 +219,11 @@ func confirmClusterConfig(clusterConfig *clusterconfig.ClusterConfig, awsCreds *
 	if clusterConfig.LogGroup != defaultCC.LogGroup {
 		items = append(items, table.KV{K: "log group", V: clusterConfig.LogGroup})
 	}
+
+	if clusterConfig.InstanceVolumeSize != defaultCC.InstanceVolumeSize {
+		items = append(items, table.KV{K: "instance volume size", V: clusterConfig.InstanceVolumeSize})
+	}
+
 	if clusterConfig.Telemetry != defaultCC.Telemetry {
 		items = append(items, table.KV{K: "telemetry", V: clusterConfig.Telemetry})
 	}

docs/cluster/config.md

@@ -34,6 +34,9 @@ log_group: cortex
 # Name of the EKS cluster Cortex will create
 cluster_name: cortex
 
+# Volume size for each instance in the cluster (in GiB)
+instance_volume_size: 50
+
 # Flag to enable collection of anonymous usage stats and error reports
 telemetry: true
 

manager/eks.yaml

@@ -23,6 +23,7 @@ metadata:
 nodeGroups:
   - name: ng-1
     instanceType: $CORTEX_INSTANCE_TYPE
+    volumeSize: $CORTEX_INSTANCE_VOLUME_SIZE
     minSize: $CORTEX_MIN_INSTANCES
     maxSize: $CORTEX_MAX_INSTANCES
     desiredCapacity: $CORTEX_MIN_INSTANCES

pkg/lib/clusterconfig/clusterconfig.go

@@ -37,6 +37,7 @@ type ClusterConfig struct {
 	Region                 string  `json:"region" yaml:"region"`
 	Bucket                 string  `json:"bucket" yaml:"bucket"`
 	LogGroup               string  `json:"log_group" yaml:"log_group"`
+	InstanceVolumeSize     int64   `json:"instance_volume_size" yaml:"instance_volume_size"`
 	Telemetry              bool    `json:"telemetry" yaml:"telemetry"`
 	ImagePredictorServe    string  `json:"image_predictor_serve" yaml:"image_predictor_serve"`
 	ImagePredictorServeGPU string  `json:"image_predictor_serve_gpu" yaml:"image_predictor_serve_gpu"`
@@ -111,6 +112,14 @@ var Validation = &cr.StructValidation{
 				Default: "cortex",
 			},
 		},
+		{
+			StructField: "InstanceVolumeSize",
+			Int64Validation: &cr.Int64Validation{
+				Default:              50,
+				GreaterThanOrEqualTo: pointer.Int64(20), // large enough to fit docker images and any other overhead
+				LessThanOrEqualTo:    pointer.Int64(16384),
+			},
+		},
 		{
 			StructField: "Telemetry",
 			BoolValidation: &cr.BoolValidation{
@@ -373,6 +382,7 @@ func (cc *InternalClusterConfig) String() string {
 	items = append(items, table.KV{K: "region", V: cc.Region})
 	items = append(items, table.KV{K: "bucket", V: cc.Bucket})
 	items = append(items, table.KV{K: "log group", V: cc.LogGroup})
+	items = append(items, table.KV{K: "instance volume size", V: cc.InstanceVolumeSize})
 	items = append(items, table.KV{K: "telemetry", V: cc.Telemetry})
 	items = append(items, table.KV{K: "image_predictor_serve", V: cc.ImagePredictorServe})
 	items = append(items, table.KV{K: "image_predictor_serve_gpu", V: cc.ImagePredictorServeGPU})