add support for e2e latency for dynamic mode

pkg/fileutil/util.go

@@ -2,6 +2,7 @@ package fileutil
 
 import (
 	"os"
+	"strconv"
 	"time"
 
 	"github.com/fsnotify/fsnotify"
@@ -104,3 +105,20 @@ func StartLoadDynamicFile(filename string, callBack FileChangeCallBack, stopCh <
 		}
 	}, time.Second, stopCh)
 }
+
+func CalculateTimeDeltaFromUnixInSeconds(from, to string) (float64, error) {
+	parsedFrom, err := strconv.ParseInt(from, 10, 64)
+	if err != nil {
+		return 0, err
+	}
+
+	parsedTo, err := strconv.ParseInt(to, 10, 64)
+	if err != nil {
+		return 0, err
+	}
+
+	timeFrom := time.Unix(parsedFrom, 0).UTC()
+	timeTo := time.Unix(parsedTo, 0).UTC()
+
+	return timeTo.Sub(timeFrom).Seconds(), nil
+}

pkg/fileutil/util_test.go

@@ -122,3 +122,34 @@ func TestDeleteDynamicFile(t *testing.T) {
 	}
 	testA.mutex.Unlock()
 }
+
+func TestCalculateTimeDeltaFromUnixInSeconds(t *testing.T) {
+	type args struct {
+		from string
+		to   string
+	}
+	cases := []struct {
+		input args
+		want  float64
+	}{
+		{
+			args{"1706648530", "1706648539"},
+			9.0,
+		},
+		{
+			args{"1706648520", "1706648539"},
+			19.0,
+		},
+	}
+
+	for _, c := range cases {
+		out, err := CalculateTimeDeltaFromUnixInSeconds(c.input.from, c.input.to)
+		if err != nil {
+			t.Errorf("error is not expected")
+		}
+
+		if out != c.want {
+			t.Errorf("unexpected result: got %v but expected %v", out, c.want)
+		}
+	}
+}

pkg/mapper/dynamicfile/dynamicfile.go

@@ -3,13 +3,17 @@ package dynamicfile
 import (
 	"encoding/json"
 	"fmt"
+	"strconv"
 	"strings"
 	"sync"
+	"time"
 
 	"github.com/sirupsen/logrus"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/arn"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/config"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/errutil"
+	"sigs.k8s.io/aws-iam-authenticator/pkg/fileutil"
+	"sigs.k8s.io/aws-iam-authenticator/pkg/metrics"
 )
 
 type DynamicFileMapStore struct {
@@ -21,9 +25,25 @@ type DynamicFileMapStore struct {
 	filename                  string
 	userIDStrict              bool
 	usernamePrefixReserveList []string
+
+	skip bool
+}
+
+// Meta is the collection of fields which should be included in all top-level
+// objects which are propagated for dynamic file mode
+type Meta struct {
+	APIVersion string `json:"ApiVersion"`
+	// Time that the object takes from update time to load time
+	LastUpdatedDateTime string `json:"LastUpdatedDateTime"`
+	// Version is the version number of the update
+	Version string `json:"Version"`
+	// ClusterID is the id of the cluster
+	ClusterID string `json:"ClusterId"`
 }
 
 type DynamicFileData struct {
+	Meta
+
 	// RoleMappings is a list of mappings from AWS IAM Role to
 	// Kubernetes username + groups.
 	RoleMappings []config.RoleMapping `json:"mapRoles"`
@@ -48,6 +68,7 @@ func NewDynamicFileMapStore(cfg config.Config) (*DynamicFileMapStore, error) {
 	ms := DynamicFileMapStore{}
 	ms.filename = cfg.DynamicFilePath
 	ms.userIDStrict = cfg.DynamicFileUserIDStrict
+	ms.skip = true
 	return &ms, nil
 }
 
@@ -165,6 +186,24 @@ func (ms *DynamicFileMapStore) CallBackForFileLoad(dynamicContent []byte) error
 		return err
 	}
 	ms.saveMap(userMappings, roleMappings, awsAccounts)
+
+	// emit load latency and if first time, skip emit and set to false to avoid instance bounce issue
+	if ms.skip {
+		ms.skip = false
+	} else {
+		latency, err := fileutil.CalculateTimeDeltaFromUnixInSeconds(dynamicFileData.LastUpdatedDateTime, strconv.FormatInt(time.Now().Unix(), 10))
+		if err != nil {
+			return fmt.Errorf("error parsing latency for dynamic file: %v", err)
+		}
+		metrics.Get().E2ELatency.WithLabelValues("dynamic_file").Observe(latency)
+		logrus.WithFields(logrus.Fields{
+			"ClusterId": dynamicFileData.ClusterID,
+			"Version":   dynamicFileData.Version,
+			"Type":      "dynamic_file",
+			"Latency":   latency,
+		}).Infof("logging latency metric")
+	}
+
 	return nil
 }
 

pkg/mapper/dynamicfile/dynamicfile_test.go

@@ -7,9 +7,11 @@ import (
 	"time"
 
 	"github.com/google/go-cmp/cmp"
+	"github.com/prometheus/client_golang/prometheus"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/config"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/errutil"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/fileutil"
+	"sigs.k8s.io/aws-iam-authenticator/pkg/metrics"
 	"sigs.k8s.io/aws-iam-authenticator/pkg/token"
 )
 
@@ -18,6 +20,11 @@ var (
 	testRole = config.RoleMapping{RoleARN: "arn:aws:iam::012345678912:role/computer", Username: "computer", Groups: []string{"system:nodes"}}
 )
 
+func TestMain(m *testing.M) {
+	metrics.InitMetrics(prometheus.NewRegistry())
+	m.Run()
+}
+
 func makeStore(users map[string]config.UserMapping, roles map[string]config.RoleMapping, filename string, userIDStrict bool) DynamicFileMapStore {
 	ms := DynamicFileMapStore{
 		users:        users,
@@ -96,6 +103,10 @@ func TestAWSAccount(t *testing.T) {
 
 var origFileContent = `
 {
+  "ApiVersion": "1",
+  "Version": "1",
+  "LastUpdatedDateTime": "12345678",
+  "ClusterId": "000000000098",
   "mapRoles": [
     {
       "rolearn": "arn:aws:iam::000000000098:role/KubernetesAdmin",
@@ -133,6 +144,10 @@ var origFileContent = `
 
 var updatedFileContent = `
 {
+  "ApiVersion": "1",
+  "Version": "1",
+  "LastUpdatedDateTime": "12345678",
+  "ClusterId": "000000000098",
   "mapRoles": [
     {
       "rolearn": "arn:aws:iam::000000000098:role/KubernetesAdmin",

pkg/metrics/metrics.go

@@ -42,6 +42,7 @@ type Metrics struct {
 	StsResponses                 *prometheus.CounterVec
 	DynamicFileFailures          prometheus.Counter
 	StsThrottling                prometheus.Counter
+	E2ELatency                   *prometheus.HistogramVec
 }
 
 func createMetrics(reg prometheus.Registerer) Metrics {
@@ -98,5 +99,14 @@ func createMetrics(reg prometheus.Registerer) Metrics {
 				Help:      "Number of EC2 describe instances calls.",
 			},
 		),
+		E2ELatency: factory.NewHistogramVec(
+			prometheus.HistogramOpts{
+				Name:      "dynamic_e2e_latency_seconds",
+				Namespace: Namespace,
+				Help:      "End to end latency in seconds partitioned by type.",
+				Buckets:   prometheus.ExponentialBuckets(2, 2, 8),
+			},
+			[]string{"type"},
+		),
 	}
 }

pkg/server/server.go

@@ -24,6 +24,7 @@ import (
 	"log"
 	"net/http"
 	"regexp"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -75,6 +76,7 @@ type handler struct {
 	ec2Provider      ec2provider.EC2Provider
 	clusterID        string
 	backendMapper    BackendMapper
+	skip             bool
 	scrubbedAccounts []string
 	cfg              config.Config
 }
@@ -211,6 +213,7 @@ func (c *Server) getHandler(backendMapper BackendMapper, ec2DescribeQps int, ec2
 		backendMapper:    backendMapper,
 		scrubbedAccounts: c.Config.ScrubbedAWSAccounts,
 		cfg:              c.Config,
+		skip:             true,
 	}
 
 	h.HandleFunc("/authenticate", h.authenticateEndpoint)
@@ -513,6 +516,24 @@ func (h *handler) CallBackForFileLoad(dynamicContent []byte) error {
 	} else {
 		logrus.Infof("BackendMode dynamic file got changed, but same with current mode, skip rebuild mapper")
 	}
+
+	// emit load latency and if first time, skip emit and set to false to avoid instance bounce issue
+	if h.skip {
+		h.skip = false
+	} else {
+		latency, err := fileutil.CalculateTimeDeltaFromUnixInSeconds(backendModes.LastUpdatedDateTime, strconv.FormatInt(time.Now().Unix(), 10))
+		if err != nil {
+			return fmt.Errorf("error parsing latency for dynamic backend mode file: %v", err)
+		}
+		metrics.Get().E2ELatency.WithLabelValues("dynamic_backend_mode").Observe(latency)
+		logrus.WithFields(logrus.Fields{
+			"ClusterId": backendModes.ClusterID,
+			"Version":   backendModes.Version,
+			"Type":      "dynamic_backend_mode",
+			"Latency":   latency,
+		}).Infof("logging latency metric")
+	}
+
 	return nil
 }
 

pkg/server/types.go

@@ -39,7 +39,20 @@ type BackendMapper struct {
 	currentModes string
 }
 
+// Meta is the collection of fields which should be included in all top-level
+// objects which are propagated for dynamic file mode
+type Meta struct {
+	APIVersion string `json:"ApiVersion"`
+	// Time that the object takes from update time to load time
+	LastUpdatedDateTime string `json:"LastUpdatedDateTime"`
+	// Version is the version number of the update
+	Version string `json:"Version"`
+	// ClusterID is the id of the cluster
+	ClusterID string `json:"ClusterId"`
+}
+
 // AccessConfig represents the configuration format for cluster access config via backend mode.
 type BackendModeConfig struct {
+	Meta
 	BackendMode string `json:"backendMode"`
 }