Add New Metrics For Where Customer Are Using The Agent (#913)

aws · Oct 23, 2023 · 64d9a5c · 64d9a5c
1 parent b6c6bfa
commit 64d9a5c
Show file tree

Hide file tree

Showing 41 changed files with 183 additions and 22 deletions.
diff --git a/cmd/config-downloader/downloader.go b/cmd/config-downloader/downloader.go
@@ -143,7 +143,7 @@ func main() {
 
 	mode = sdkutil.DetectAgentMode(mode)
 
-	region = util.DetectRegion(mode, cc.CredentialsMap())
+	region, _ = util.DetectRegion(mode, cc.CredentialsMap())
 
 	if region == "" && downloadLocation != locationDefault {
 		fmt.Println("Unable to determine aws-region.")

diff --git a/handlers/agentinfo/info.go b/handlers/agentinfo/info.go
@@ -30,12 +30,22 @@ import (
 	"github.com/aws/amazon-cloudwatch-agent/cfg/envconfig"
 	"github.com/aws/amazon-cloudwatch-agent/internal/util/collections"
 	"github.com/aws/amazon-cloudwatch-agent/receiver/adapter"
+	translatorConfig "github.com/aws/amazon-cloudwatch-agent/translator/config"
 )
 
 const (
 	versionFilename = "CWAGENT_VERSION"
 	unknownVersion  = "Unknown"
 	updateInterval  = time.Minute
+	// region types
+	AgentConfigJson = "ACJ"
+	CredsMap        = "CM"
+	EC2Metadata     = "EC2M"
+	ECSMetadata     = "ECSM"
+	RegionNotFound  = "RNF"
+	ModeEC2         = "EC2"
+	ModeOnPrem      = "OP"
+	ModeWithIRSA    = "WI"
 )
 
 var (
@@ -50,10 +60,12 @@ var (
 	id                      = uuid.NewString()
 	sharedConfigFallback    atomic.Bool
 	imdsFallbackSucceed     atomic.Bool
+	isRunningAsRoot         = defaultIsRunningAsRoot
+	runInContainer          *int
+	regionType              *string
+	mode                    *string
 )
 
-var isRunningAsRoot = defaultIsRunningAsRoot
-
 type AgentInfo interface {
 	RecordOpData(time.Duration, int, error)
 	StatsHeader() string
@@ -77,15 +89,21 @@ type agentStats struct {
 	StatusCode           *int     `json:"code,omitempty"`
 	SharedConfigFallback *int     `json:"scfb,omitempty"`
 	ImdsFallbackSucceed  *int     `json:"ifs,omitempty"`
+	RunInContainer       *int     `json:"ric,omitempty"`
+	RegionType           *string  `json:"rt,omitempty"`
+	Mode                 *string  `json:"m,omitempty"`
 }
 
-func New(groupName string) AgentInfo {
-	return newAgentInfo(groupName)
+func New(groupName string, regionType string, mode string) AgentInfo {
+	return newAgentInfo(groupName, regionType, mode)
 }
 
-func newAgentInfo(groupName string) *agentInfo {
+func newAgentInfo(groupName string, regionTypeInput string, modeInput string) *agentInfo {
 	ai := new(agentInfo)
 	ai.userAgent = getUserAgent(groupName, fullVersion, receivers, processors, exporters, isUsageDataEnabled())
+	runInContainer = runInContainerFunc()
+	regionType = aws.String(regionTypeInput)
+	mode = aws.String(modeInput)
 	if isUsageDataEnabled() {
 		ai.proc, _ = process.NewProcess(int32(os.Getpid()))
 		if ai.proc == nil {
@@ -96,6 +114,9 @@ func newAgentInfo(groupName string) *agentInfo {
 			MemoryBytes:         ai.memoryBytes(),
 			FileDescriptorCount: ai.fileDescriptorCount(),
 			ThreadCount:         ai.threadCount(),
+			RunInContainer:      runInContainer,
+			RegionType:          regionType,
+			Mode:                mode,
 		}
 		ai.statsHeader = getAgentStats(stats)
 		ai.nextUpdate = time.Now().Add(updateInterval)
@@ -126,6 +147,9 @@ func (ai *agentInfo) RecordOpData(latency time.Duration, payloadBytes int, err e
 		stats.ThreadCount = ai.threadCount()
 		stats.SharedConfigFallback = getSharedConfigFallback()
 		stats.ImdsFallbackSucceed = succeedImdsFallback()
+		stats.RunInContainer = runInContainer
+		stats.RegionType = regionType
+		stats.Mode = mode
 		ai.nextUpdate = now.Add(updateInterval)
 	}
 
@@ -330,3 +354,10 @@ func getSharedConfigFallback() *int {
 func SetImdsFallbackSucceed() {
 	imdsFallbackSucceed.Store(true)
 }
+
+func runInContainerFunc() *int {
+	if os.Getenv(translatorConfig.RUN_IN_CONTAINER) == translatorConfig.RUN_IN_CONTAINER_TRUE {
+		return aws.Int(1)
+	}
+	return aws.Int(0)
+}
diff --git a/handlers/agentinfo/info_test.go b/handlers/agentinfo/info_test.go
@@ -29,15 +29,15 @@ import (
 )
 
 func TestNew(t *testing.T) {
-	ai := New("")
+	ai := New("", "", "")
 	expectedUserAgentRegex := `^CWAgent/Unknown \(.*\) ` +
 		`ID/[0-9a-fA-F]{8}\b-[0-9a-fA-F]{4}\b-[0-9a-fA-F]{4}\b-[0-9a-fA-F]{4}\b-[0-9a-fA-F]{12}$`
 
 	assert.Regexp(t, regexp.MustCompile(expectedUserAgentRegex), ai.UserAgent())
 }
 
 func TestRecordOpData(t *testing.T) {
-	ai := newAgentInfo("")
+	ai := newAgentInfo("", "", "")
 
 	stats := ai.StatsHeader()
 	actual := agentStats{}
@@ -124,8 +124,20 @@ func TestGetAgentStats(t *testing.T) {
 		PayloadBytes:        aws.Int(5678),
 		StatusCode:          aws.Int(200),
 		ImdsFallbackSucceed: aws.Int(1),
+		RunInContainer:      aws.Int(0),
+		RegionType:          aws.String(EC2Metadata),
+		Mode:                aws.String(ModeWithIRSA),
 	}
 
+	assert.Equal(t, "\"cpu\":1.2,\"mem\":123,\"fd\":456,\"th\":789,\"lat\":1234,\"load\":5678,\"code\":200,\"ifs\":1,\"ric\":0,\"rt\":\"EC2M\",\"m\":\"WI\"", getAgentStats(stats))
+
+	stats.Mode = nil
+	assert.Equal(t, "\"cpu\":1.2,\"mem\":123,\"fd\":456,\"th\":789,\"lat\":1234,\"load\":5678,\"code\":200,\"ifs\":1,\"ric\":0,\"rt\":\"EC2M\"", getAgentStats(stats))
+
+	stats.RegionType = nil
+	assert.Equal(t, "\"cpu\":1.2,\"mem\":123,\"fd\":456,\"th\":789,\"lat\":1234,\"load\":5678,\"code\":200,\"ifs\":1,\"ric\":0", getAgentStats(stats))
+
+	stats.RunInContainer = nil
 	assert.Equal(t, "\"cpu\":1.2,\"mem\":123,\"fd\":456,\"th\":789,\"lat\":1234,\"load\":5678,\"code\":200,\"ifs\":1", getAgentStats(stats))
 
 	stats.ImdsFallbackSucceed = nil

diff --git a/plugins/outputs/cloudwatch/cloudwatch.go b/plugins/outputs/cloudwatch/cloudwatch.go
@@ -81,7 +81,7 @@ func (c *CloudWatch) Capabilities() consumer.Capabilities {
 }
 
 func (c *CloudWatch) Start(_ context.Context, host component.Host) error {
-	c.agentInfo = agentinfo.New("")
+	c.agentInfo = agentinfo.New("", c.config.RegionType, c.config.Mode)
 	c.publisher, _ = publisher.NewPublisher(
 		publisher.NewNonBlockingFifoQueue(metricChanBufferSize),
 		maxConcurrentPublisher,

diff --git a/plugins/outputs/cloudwatch/cloudwatch_test.go b/plugins/outputs/cloudwatch/cloudwatch_test.go
@@ -382,7 +382,7 @@ func newCloudWatchClient(
 			MaxDatumsPerCall:   defaultMaxDatumsPerCall,
 			MaxValuesPerDatum:  defaultMaxValuesPerDatum,
 		},
-		agentInfo: agentinfo.New(""),
+		agentInfo: agentinfo.New("", "", ""),
 	}
 	cloudwatch.startRoutines()
 	return cloudwatch

diff --git a/plugins/outputs/cloudwatch/config.go b/plugins/outputs/cloudwatch/config.go
@@ -11,13 +11,15 @@ import (
 	"go.opentelemetry.io/collector/component"
 )
 
-// Config represent a configuration for the CloudWatch logs exporter.
+// Config represent a configuration for the CloudWatch metrics exporter.
 type Config struct {
 	Region                   string          `mapstructure:"region"`
 	EndpointOverride         string          `mapstructure:"endpoint_override,omitempty"`
 	AccessKey                string          `mapstructure:"access_key,omitempty"`
 	SecretKey                string          `mapstructure:"secret_key,omitempty"`
 	RoleARN                  string          `mapstructure:"role_arn,omitempty"`
+	RegionType               string          `mapstructure:"region_type,omitempty"`
+	Mode                     string          `mapstructure:"mode,omitempty"`
 	Profile                  string          `mapstructure:"profile,omitempty"`
 	SharedCredentialFilename string          `mapstructure:"shared_credential_file,omitempty"`
 	Token                    string          `mapstructure:"token,omitempty"`

diff --git a/plugins/outputs/cloudwatchlogs/cloudwatchlogs.go b/plugins/outputs/cloudwatchlogs/cloudwatchlogs.go
@@ -31,7 +31,7 @@ const (
 	LogEntryField     = "value"
 
 	defaultFlushTimeout = 5 * time.Second
-	eventHeaderSize     = 26
+	eventHeaderSize     = 200
 	truncatedSuffix     = "[Truncated...]"
 	msgSizeLimit        = 256*1024 - eventHeaderSize
 
@@ -43,6 +43,8 @@ const (
 
 type CloudWatchLogs struct {
 	Region           string `toml:"region"`
+	RegionType       string `toml:"region_type"`
+	Mode             string `toml:"mode"`
 	EndpointOverride string `toml:"endpoint_override"`
 	AccessKey        string `toml:"access_key"`
 	SecretKey        string `toml:"secret_key"`
@@ -133,7 +135,7 @@ func (c *CloudWatchLogs) getDest(t Target) *cwDest {
 			Logger:   configaws.SDKLogger{},
 		},
 	)
-	agentInfo := agentinfo.New(t.Group)
+	agentInfo := agentinfo.New(t.Group, c.RegionType, c.Mode)
 	client.Handlers.Build.PushBackNamed(handlers.NewRequestCompressionHandler([]string{"PutLogEvents"}))
 	client.Handlers.Build.PushBackNamed(handlers.NewCustomHeaderHandler("User-Agent", agentInfo.UserAgent()))
 	client.Handlers.Build.PushBackNamed(handlers.NewDynamicCustomHeaderHandler("X-Amz-Agent-Stats", agentInfo.StatsHeader))

diff --git a/plugins/outputs/cloudwatchlogs/pusher_test.go b/plugins/outputs/cloudwatchlogs/pusher_test.go
@@ -765,6 +765,6 @@ func TestResendWouldStopAfterExhaustedRetries(t *testing.T) {
 
 func testPreparation(retention int, s *svcMock, flushTimeout time.Duration, retryDuration time.Duration) (chan struct{}, *pusher) {
 	stop := make(chan struct{})
-	p := NewPusher(Target{"G", "S", retention}, s, flushTimeout, retryDuration, models.NewLogger("cloudwatchlogs", "test", ""), stop, &wg, agentinfo.New(""))
+	p := NewPusher(Target{"G", "S", retention}, s, flushTimeout, retryDuration, models.NewLogger("cloudwatchlogs", "test", ""), stop, &wg, agentinfo.New("", "", ""))
 	return stop, p
 }
diff --git a/translator/context/context.go b/translator/context/context.go
@@ -7,6 +7,7 @@ import (
 	"log"
 	"os"
 
+	"github.com/aws/amazon-cloudwatch-agent/handlers/agentinfo"
 	"github.com/aws/amazon-cloudwatch-agent/translator/config"
 )
 
@@ -41,6 +42,7 @@ type Context struct {
 	multiConfig         string
 	outputTomlFilePath  string
 	mode                string
+	shortMode           string
 	credentials         map[string]string
 	proxy               map[string]string
 	ssl                 map[string]string
@@ -96,6 +98,10 @@ func (ctx *Context) Mode() string {
 	return ctx.mode
 }
 
+func (ctx *Context) ShortMode() string {
+	return ctx.shortMode
+}
+
 func (ctx *Context) Credentials() map[string]string {
 	return ctx.credentials
 }
@@ -112,12 +118,16 @@ func (ctx *Context) SetMode(mode string) {
 	switch mode {
 	case config.ModeEC2:
 		ctx.mode = config.ModeEC2
+		ctx.shortMode = agentinfo.ModeEC2
 	case config.ModeOnPrem:
 		ctx.mode = config.ModeOnPrem
+		ctx.shortMode = agentinfo.ModeOnPrem
 	case config.ModeOnPremise:
 		ctx.mode = config.ModeOnPremise
+		ctx.shortMode = agentinfo.ModeOnPrem
 	case config.ModeWithIRSA:
 		ctx.mode = config.ModeWithIRSA
+		ctx.shortMode = agentinfo.ModeWithIRSA
 	default:
 		log.Panicf("Invalid mode %s. Valid mode values are %s, %s, %s and %s.", mode, config.ModeEC2, config.ModeOnPrem, config.ModeOnPremise, config.ModeWithIRSA)
 	}

diff --git a/translator/tocwconfig/sampleConfig/advanced_config_darwin.yaml b/translator/tocwconfig/sampleConfig/advanced_config_darwin.yaml
@@ -6,6 +6,8 @@ exporters:
     max_values_per_datum: 150
     namespace: CWAgent
     region: us-west-2
+    region_type: "ACJ"
+    mode: "EC2"
     resource_to_telemetry_conversion:
       enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/advanced_config_linux.yaml b/translator/tocwconfig/sampleConfig/advanced_config_linux.yaml
@@ -6,6 +6,8 @@ exporters:
         max_values_per_datum: 150
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/advanced_config_windows.yaml b/translator/tocwconfig/sampleConfig/advanced_config_windows.yaml
@@ -6,6 +6,8 @@ exporters:
         max_values_per_datum: 150
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/basic_config_linux.yaml b/translator/tocwconfig/sampleConfig/basic_config_linux.yaml
@@ -6,6 +6,8 @@ exporters:
         max_values_per_datum: 150
         namespace: CWAgent
         region: us-east-1
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/basic_config_windows.yaml b/translator/tocwconfig/sampleConfig/basic_config_windows.yaml
@@ -6,6 +6,8 @@ exporters:
         max_values_per_datum: 150
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/collectd_config_linux.yaml b/translator/tocwconfig/sampleConfig/collectd_config_linux.yaml
@@ -6,6 +6,8 @@ exporters:
         max_values_per_datum: 150
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/complete_darwin_config.conf b/translator/tocwconfig/sampleConfig/complete_darwin_config.conf
@@ -126,4 +126,6 @@
     force_flush_interval = "60s"
     log_stream_name = "LOG_STREAM_NAME"
     region = "us-west-2"
+    region_type = "ACJ"
+    mode = "EC2"
     role_arn = "log_role_arn_value_test"
diff --git a/translator/tocwconfig/sampleConfig/complete_darwin_config.yaml b/translator/tocwconfig/sampleConfig/complete_darwin_config.yaml
@@ -7,6 +7,8 @@ exporters:
         max_values_per_datum: 5000
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
         role_arn: metrics_role_arn_value_test

diff --git a/translator/tocwconfig/sampleConfig/complete_linux_config.conf b/translator/tocwconfig/sampleConfig/complete_linux_config.conf
@@ -126,4 +126,6 @@
     force_flush_interval = "60s"
     log_stream_name = "LOG_STREAM_NAME"
     region = "us-west-2"
+    region_type = "ACJ"
+    mode = "EC2"
     role_arn = "log_role_arn_value_test"
diff --git a/translator/tocwconfig/sampleConfig/complete_linux_config.yaml b/translator/tocwconfig/sampleConfig/complete_linux_config.yaml
@@ -10,6 +10,8 @@ exporters:
         max_values_per_datum: 5000
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
         role_arn: metrics_role_arn_value_test

diff --git a/translator/tocwconfig/sampleConfig/complete_windows_config.conf b/translator/tocwconfig/sampleConfig/complete_windows_config.conf
@@ -144,4 +144,6 @@
     force_flush_interval = "60s"
     log_stream_name = "LOG_STREAM_NAME"
     region = "us-west-2"
+    region_type = "ACJ"
+    mode = "EC2"
     role_arn = "log_role_arn_value_test"
diff --git a/translator/tocwconfig/sampleConfig/complete_windows_config.yaml b/translator/tocwconfig/sampleConfig/complete_windows_config.yaml
@@ -7,6 +7,8 @@ exporters:
         max_values_per_datum: 5000
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
         role_arn: metrics_role_arn_value_test

diff --git a/translator/tocwconfig/sampleConfig/delta_config_linux.yaml b/translator/tocwconfig/sampleConfig/delta_config_linux.yaml
@@ -6,6 +6,8 @@ exporters:
       max_values_per_datum: 150
       namespace: CWAgent
       region: us-east-1
+      region_type: "ACJ"
+      mode: "EC2"
       resource_to_telemetry_conversion:
         enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/delta_net_config_linux.yaml b/translator/tocwconfig/sampleConfig/delta_net_config_linux.yaml
@@ -6,6 +6,8 @@ exporters:
       max_values_per_datum: 150
       namespace: CWAgent
       region: us-east-1
+      region_type: "ACJ"
+      mode: "EC2"
       resource_to_telemetry_conversion:
         enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/drop_origin_linux.yaml b/translator/tocwconfig/sampleConfig/drop_origin_linux.yaml
@@ -11,6 +11,8 @@ exporters:
         max_values_per_datum: 150
         namespace: CWAgent
         region: us-west-2
+        region_type: "ACJ"
+        mode: "EC2"
         resource_to_telemetry_conversion:
             enabled: true
 extensions: {}

diff --git a/translator/tocwconfig/sampleConfig/ignore_append_dimensions.yaml b/translator/tocwconfig/sampleConfig/ignore_append_dimensions.yaml
@@ -6,6 +6,8 @@ exporters:
     max_values_per_datum: 150
     namespace: CWAgent
     region: us-east-1
+    region_type: "ACJ"
+    mode: "EC2"
     resource_to_telemetry_conversion:
       enabled: true
 extensions: {}