removed manifest complexity

Author: annapoornanarayan

test/cases/common/config.go

@@ -0,0 +1,24 @@
+//go:build e2e
+
+package common
+
+import (
+	"flag"
+	"github.com/octago/sflags/gen/gpflag"
+	"github.com/spf13/pflag"
+)
+
+// for adding common flags using gpflag
+var MetricDimensions = make(map[string]string)
+
+func InitFlags(config interface{}) {
+	flags, err := gpflag.Parse(config)
+	if err != nil {
+		panic(err)
+	}
+	flags.StringToStringVar(&MetricDimensions, "metricDimensions", nil, "Metric dimensions as key=value pairs")
+
+	flags.VisitAll(func(pf *pflag.Flag) {
+		flag.CommandLine.Var(pf.Value, pf.Name, pf.Usage)
+	})
+}

test/cases/nvidia-training/main_test.go

@@ -6,6 +6,7 @@ import (
 	"context"
 	_ "embed"
 	"fmt"
+	"github.com/aws/aws-k8s-tester/test/cases/common"
 	"log"
 	"os"
 	"os/signal"
@@ -36,7 +37,7 @@ func TestMain(m *testing.M) {
 	testenv = env.NewWithConfig(cfg).WithContext(ctx)
 
 	// Render CloudWatch Agent manifest with dynamic dimensions
-	renderedCloudWatchAgentManifest, err := manifests.RenderCloudWatchAgentManifest(testConfig.MetricDimensions)
+	renderedCloudWatchAgentManifest, err := manifests.RenderCloudWatchAgentManifest(common.MetricDimensions)
 	if err != nil {
 		log.Printf("Warning: failed to render CloudWatch Agent manifest: %v", err)
 	}

test/cases/nvidia-training/vars.go

@@ -3,17 +3,14 @@
 package training
 
 import (
-	"flag"
-	"github.com/octago/sflags/gen/gpflag"
-	"github.com/spf13/pflag"
+	"github.com/aws/aws-k8s-tester/test/cases/common"
 	"sigs.k8s.io/e2e-framework/pkg/env"
 )
 
 type Config struct {
-	BertTrainingImage string   `flag:"bertTrainingImage" desc:"Docker image used for BERT training workload"`
-	EfaEnabled        bool     `flag:"efaEnabled" desc:"Enable Elastic Fabric Adapter (EFA)"`
-	NodeType          string   `flag:"nodeType" desc:"Instance type for cluster nodes"`
-	MetricDimensions  []string `flag:"metricDimensions" desc:"Metric dimensions as comma-separated key=value pairs"`
+	BertTrainingImage string `flag:"bertTrainingImage" desc:"Docker image used for BERT training workload"`
+	EfaEnabled        bool   `flag:"efaEnabled" desc:"Enable Elastic Fabric Adapter (EFA)"`
+	NodeType          string `flag:"nodeType" desc:"Instance type for cluster nodes"`
 }
 
 // Shared global variables
@@ -27,11 +24,5 @@ var (
 )
 
 func init() {
-	flags, err := gpflag.Parse(testConfig)
-	if err != nil {
-		panic(err)
-	}
-	flags.VisitAll(func(pf *pflag.Flag) {
-		flag.CommandLine.Var(pf.Value, pf.Name, pf.Usage)
-	})
+	common.InitFlags(testConfig)
 }

test/manifests/assets/cloudwatch-agent.yaml

@@ -18,7 +18,7 @@ data:
                 {
                   "source_labels": ["job"],
                   "label_matcher": "dcgm-exporter",
-                  "dimensions": [[{{- if .MetricDimensions}}{{$first := true}}{{range $key, $value := .MetricDimensions}}{{if not $first}}, {{end}}"{{$key}}"{{$first = false}}{{end}}{{- end}}]],
+                  "dimensions": [[{{.DimensionKeys}}]],
                   "metric_selectors": [
                     "^DCGM_FI_DEV_GPU_UTIL$",
                     "^DCGM_FI_DEV_MEM_COPY_UTIL$",
@@ -53,11 +53,9 @@ data:
             - dcgm-exporter.kube-system.svc.cluster.local:9400
         metrics_path: /metrics
         metric_relabel_configs:
-{{- if .MetricDimensions}}
 {{- range $key, $value := .MetricDimensions}}
           - {action: replace, target_label: {{$key}}, replacement: '{{$value}}'}
 {{- end}}
-{{- end}}
 ---
 apiVersion: apps/v1
 kind: DaemonSet

test/manifests/raw.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	_ "embed"
 	"fmt"
+	"html/template"
 	"log"
 	"strings"
 	"time"
@@ -37,34 +38,17 @@ var (
 	CloudWatchAgentManifest []byte
 )
 
-// GetMetricDimensions parses CW metric dimensions from key=value pairs
-func GetMetricDimensions(metricDimensions []string) (map[string]string, error) {
-	dimensions := make(map[string]string)
-	for _, pair := range metricDimensions {
-		kv := strings.Split(strings.TrimSpace(pair), "=")
-		if len(kv) != 2 {
-			return nil, fmt.Errorf("invalid dimension: %q", pair)
-		}
-		key, value := strings.TrimSpace(kv[0]), strings.TrimSpace(kv[1])
-		if key == "" || value == "" {
-			return nil, fmt.Errorf("invalid dimension: %q", pair)
-		}
-		dimensions[key] = value
-	}
-	return dimensions, nil
-}
-
 // RenderCloudWatchAgentManifest renders the CloudWatch Agent manifest with dynamic dimensions
-func RenderCloudWatchAgentManifest(metricDimensionsSlice []string) ([]byte, error) {
-	metricDimensions, err := GetMetricDimensions(metricDimensionsSlice)
-	if err != nil {
-		return nil, fmt.Errorf("failed to parse metric dimensions: %w", err)
-	}
-	renderedManifest, err := fwext.RenderManifests(CloudWatchAgentManifest, map[string]interface{}{"MetricDimensions": metricDimensions})
-	if err != nil {
-		return nil, fmt.Errorf("failed to render CloudWatch Agent manifest: %w", err)
+func RenderCloudWatchAgentManifest(metricDimensions map[string]string) ([]byte, error) {
+	var keys []string
+	for key := range metricDimensions {
+		keys = append(keys, `"`+key+`"`)
 	}
-	return renderedManifest, nil
+	dimensionsStr := strings.Join(keys, ", ")
+	return fwext.RenderManifests(CloudWatchAgentManifest, map[string]interface{}{
+		"MetricDimensions": metricDimensions,
+		"DimensionKeys":    template.HTML(dimensionsStr),
+	})
 }
 
 // DeployDaemonSet returns a function to deploy and wait for a DaemonSet to be ready