Remove KeepaliveTimeout from entities (#2045)

Signed-off-by: Simon Plourde <simon@sensu.io>

CHANGELOG.md

@@ -12,23 +12,28 @@ Versioning](http://semver.org/spec/v2.0.0.html).
 
 ### Changed
 - Changed sensuctl title colour to use terminal's configured default for bold
-  text.
+text.
 - The backend no longer forcibly binds to localhost.
+- Keepalive intervals and timeouts are now configured in the check object of
+keepalive events.
 - The sensu-agent binary is now located at ./cmd/sensu-agent.
 - Sensuctl no longer uses auto text wrapping.
 
 ### Fixed
 - Fixes a bug in `sensuctl cluster health` so the correct error is handled.
 - Fixed a bug where assets could not extract git tarballs.
 - Fixed a bug where assets would not install if given cache directory was a
-  relative path.
+relative path.
 - Fixed a bug where an agent's collection of system information could delay
-  sending of keepalive messages.
+sending of keepalive messages.
 - Fixed a bug in nagios perfdata parsing.
 - Etcd client URLs can now be a comma-separated list.
 - Fixed a bug where output metric format could not be unset.
 - Fixed a bug where the agent does not validate the ID at startup.
 
+### Breaking Changes
+- Removed the KeepaliveTimeout attribute from entities.
+
 ## [2.0.0-beta.4] - 2018-08-14
 
 ### Added

agent/agent.go

@@ -46,8 +46,6 @@ const (
 	DefaultEnvironment = "default"
 	// DefaultKeepaliveInterval specifies the default keepalive interval
 	DefaultKeepaliveInterval = 20
-	// DefaultKeepaliveTimeout specifies the default keepalive timeout
-	DefaultKeepaliveTimeout = 120
 	// DefaultOrganization specifies the default organization
 	DefaultOrganization = "default"
 	// DefaultPassword specifies the default password
@@ -92,10 +90,11 @@ type Config struct {
 	// start
 	ExtendedAttributes []byte
 	// KeepaliveInterval is the interval, in seconds, when agents will send a
-	// keepalive to sensu-backend. Default: 60
-	KeepaliveInterval int
+	// keepalive to sensu-backend.
+	KeepaliveInterval uint32
 	// KeepaliveTimeout is the time after which a sensu-agent is considered dead
-	// by the backend.
+	// by the backend. See DefaultKeepaliveTimeout in types package for default
+	// value.
 	KeepaliveTimeout uint32
 	// Organization sets the Agent's RBAC organization identifier
 	Organization string
@@ -143,7 +142,7 @@ func FixtureConfig() *Config {
 		CacheDir:          path.SystemCacheDir("sensu-agent"),
 		Environment:       DefaultEnvironment,
 		KeepaliveInterval: DefaultKeepaliveInterval,
-		KeepaliveTimeout:  DefaultKeepaliveTimeout,
+		KeepaliveTimeout:  types.DefaultKeepaliveTimeout,
 		Organization:      DefaultOrganization,
 		Password:          DefaultPassword,
 		Socket: &SocketConfig{
@@ -381,15 +380,19 @@ func (a *Agent) sendKeepalive() error {
 	}
 	keepalive := &types.Event{}
 
+	keepalive.Check = &types.Check{
+		Name:     "keepalive",
+		Interval: a.config.KeepaliveInterval,
+		Timeout:  a.config.KeepaliveTimeout,
+	}
 	keepalive.Entity = a.getAgentEntity()
-
 	keepalive.Timestamp = time.Now().Unix()
+
 	msgBytes, err := json.Marshal(keepalive)
 	if err != nil {
 		return err
 	}
 	msg.Payload = msgBytes
-
 	a.sendq <- msg
 
 	return nil

agent/entity.go

@@ -11,16 +11,15 @@ import (
 func (a *Agent) getAgentEntity() *types.Entity {
 	if a.entity == nil {
 		e := &types.Entity{
-			Class:            types.EntityAgentClass,
-			Deregister:       a.config.Deregister,
-			Environment:      a.config.Environment,
-			ID:               a.config.AgentID,
-			KeepaliveTimeout: a.config.KeepaliveTimeout,
-			LastSeen:         time.Now().Unix(),
-			Organization:     a.config.Organization,
-			Redact:           a.config.Redact,
-			Subscriptions:    a.config.Subscriptions,
-			User:             a.config.User,
+			Class:         types.EntityAgentClass,
+			Deregister:    a.config.Deregister,
+			Environment:   a.config.Environment,
+			ID:            a.config.AgentID,
+			LastSeen:      time.Now().Unix(),
+			Organization:  a.config.Organization,
+			Redact:        a.config.Redact,
+			Subscriptions: a.config.Subscriptions,
+			User:          a.config.User,
 		}
 
 		if a.config.DeregistrationHandler != "" {

backend/keepalived/deregisterer.go

@@ -61,7 +61,7 @@ func (adapterPtr *Deregistration) Deregister(entity *types.Entity) error {
 	if entity.Deregistration.Handler != "" {
 		deregistrationCheck := &types.Check{
 			Name:          "deregistration",
-			Interval:      entity.KeepaliveTimeout,
+			Interval:      1,
 			Subscriptions: []string{""},
 			Command:       "",
 			Handlers:      []string{entity.Deregistration.Handler},

backend/keepalived/integration_test.go

@@ -65,9 +65,9 @@ func TestKeepaliveMonitor(t *testing.T) {
 	}
 
 	entity := types.FixtureEntity("entity1")
-	entity.KeepaliveTimeout = 1
 
 	keepalive := &types.Event{
+		Check:     &types.Check{Timeout: 1},
 		Entity:    entity,
 		Timestamp: time.Now().Unix(),
 	}

backend/keepalived/keepalived.go

@@ -16,10 +16,6 @@ const (
 	// handling keepalive events.
 	DefaultHandlerCount = 10
 
-	// DefaultKeepaliveTimeout is the amount of time we consider a Keepalive
-	// valid for.
-	DefaultKeepaliveTimeout = 120 // seconds
-
 	// KeepaliveCheckName is the name of the check that is created when a
 	// keepalive timeout occurs.
 	KeepaliveCheckName = "keepalive"
@@ -217,7 +213,13 @@ func (k *Keepalived) processKeepalives() {
 			logger.WithError(err).Error("error handling entity registration")
 		}
 
-		timeout := int64(entity.KeepaliveTimeout)
+		// Retrieve the keepalive timeout or use a default value in case an older
+		// agent version was used, since entity.KeepaliveTimeout no longer exist
+		timeout := int64(types.DefaultKeepaliveTimeout)
+		if event.Check != nil {
+			timeout = int64(event.Check.Timeout)
+		}
+
 		supervisor := k.monitorFactory(k)
 		if err := supervisor.Monitor(context.TODO(), entity.ID, event, timeout); err != nil {
 			logger.WithError(err).Error("error monitoring entity")
@@ -253,13 +255,14 @@ func (k *Keepalived) handleEntityRegistration(entity *types.Entity) error {
 	return err
 }
 
-func createKeepaliveEvent(entity *types.Entity) *types.Event {
+func createKeepaliveEvent(rawEvent *types.Event) *types.Event {
 	keepaliveCheck := &types.Check{
 		Name:         KeepaliveCheckName,
-		Interval:     entity.KeepaliveTimeout,
+		Interval:     rawEvent.Check.Interval,
+		Timeout:      rawEvent.Check.Timeout,
 		Handlers:     []string{KeepaliveHandlerName},
-		Environment:  entity.Environment,
-		Organization: entity.Organization,
+		Environment:  rawEvent.Entity.Environment,
+		Organization: rawEvent.Entity.Organization,
 		Status:       1,
 		Issued:       time.Now().Unix(),
 		History: []types.CheckHistory{
@@ -271,7 +274,7 @@ func createKeepaliveEvent(entity *types.Entity) *types.Event {
 	}
 	keepaliveEvent := &types.Event{
 		Timestamp: time.Now().Unix(),
-		Entity:    entity,
+		Entity:    rawEvent.Entity,
 		Check:     keepaliveCheck,
 	}
 
@@ -281,7 +284,7 @@ func createKeepaliveEvent(entity *types.Entity) *types.Event {
 func createRegistrationEvent(entity *types.Entity) *types.Event {
 	registrationCheck := &types.Check{
 		Name:         RegistrationCheckName,
-		Interval:     entity.KeepaliveTimeout,
+		Interval:     1,
 		Handlers:     []string{RegistrationHandlerName},
 		Environment:  entity.Environment,
 		Organization: entity.Organization,
@@ -312,7 +315,7 @@ func (k *Keepalived) handleUpdate(e *types.Event) error {
 		logger.WithError(err).Error("error updating entity in store")
 		return err
 	}
-	event := createKeepaliveEvent(entity)
+	event := createKeepaliveEvent(e)
 	event.Check.Status = 0
 	return k.bus.Publish(messaging.TopicEventRaw, event)
 }
@@ -335,13 +338,13 @@ func (k *Keepalived) HandleFailure(e *types.Event) error {
 	}
 
 	// this is a real keepalive event, emit it.
-	event := createKeepaliveEvent(entity)
+	event := createKeepaliveEvent(e)
 	event.Check.Status = 1
 	if err := k.bus.Publish(messaging.TopicEventRaw, event); err != nil {
 		return err
 	}
 
 	logger.WithField("entity", entity.GetID()).Info("keepalive timed out, creating keepalive event for entity")
-	timeout := time.Now().Unix() + int64(entity.KeepaliveTimeout)
+	timeout := time.Now().Unix() + int64(e.Check.Timeout)
 	return k.store.UpdateFailingKeepalive(ctx, entity, timeout)
 }

backend/keepalived/keepalived_test.go

@@ -242,10 +242,10 @@ func TestProcessRegistration(t *testing.T) {
 }
 
 func TestCreateKeepaliveEvent(t *testing.T) {
-	entity := types.FixtureEntity("entity1")
-	keepaliveEvent := createKeepaliveEvent(entity)
+	event := types.FixtureEvent("entity1", "keepalive")
+	keepaliveEvent := createKeepaliveEvent(event)
 	assert.Equal(t, "keepalive", keepaliveEvent.Check.Name)
-	assert.Equal(t, uint32(120), keepaliveEvent.Check.Interval)
+	assert.Equal(t, uint32(60), keepaliveEvent.Check.Interval)
 	assert.Equal(t, []string{"keepalive"}, keepaliveEvent.Check.Handlers)
 	assert.Equal(t, uint32(1), keepaliveEvent.Check.Status)
 	assert.NotEqual(t, int64(0), keepaliveEvent.Check.Issued)

benchmark/cmd/main.go

@@ -11,6 +11,7 @@ import (
 
 	"github.com/google/uuid"
 	"github.com/sensu/sensu-go/agent"
+	"github.com/sensu/sensu-go/types"
 	"github.com/sensu/sensu-go/util/path"
 )
 
@@ -40,7 +41,7 @@ func main() {
 		cfg.Environment = agent.DefaultEnvironment
 		cfg.ExtendedAttributes = []byte{}
 		cfg.KeepaliveInterval = agent.DefaultKeepaliveInterval
-		cfg.KeepaliveTimeout = agent.DefaultKeepaliveTimeout
+		cfg.KeepaliveTimeout = types.DefaultKeepaliveTimeout
 		cfg.Organization = agent.DefaultOrganization
 		cfg.Password = agent.DefaultPassword
 		cfg.Socket.Host = agent.DefaultAPIHost

cmd/sensu-agent/start.go

@@ -11,6 +11,7 @@ import (
 	"syscall"
 
 	"github.com/sensu/sensu-go/agent"
+	"github.com/sensu/sensu-go/types"
 	"github.com/sensu/sensu-go/types/dynamic"
 	"github.com/sensu/sensu-go/util/path"
 	"github.com/sensu/sensu-go/util/url"
@@ -126,7 +127,7 @@ func newStartCommand() *cobra.Command {
 			cfg.DeregistrationHandler = viper.GetString(flagDeregistrationHandler)
 			cfg.Environment = viper.GetString(flagEnvironment)
 			cfg.ExtendedAttributes = []byte(viper.GetString(flagExtendedAttributes))
-			cfg.KeepaliveInterval = viper.GetInt(flagKeepaliveInterval)
+			cfg.KeepaliveInterval = uint32(viper.GetInt(flagKeepaliveInterval))
 			cfg.KeepaliveTimeout = uint32(viper.GetInt(flagKeepaliveTimeout))
 			cfg.Organization = viper.GetString(flagOrganization)
 			cfg.Password = viper.GetString(flagPassword)
@@ -228,7 +229,7 @@ func newStartCommand() *cobra.Command {
 	viper.SetDefault(flagDeregistrationHandler, "")
 	viper.SetDefault(flagEnvironment, agent.DefaultEnvironment)
 	viper.SetDefault(flagKeepaliveInterval, agent.DefaultKeepaliveInterval)
-	viper.SetDefault(flagKeepaliveTimeout, agent.DefaultKeepaliveTimeout)
+	viper.SetDefault(flagKeepaliveTimeout, types.DefaultKeepaliveTimeout)
 	viper.SetDefault(flagOrganization, agent.DefaultOrganization)
 	viper.SetDefault(flagPassword, agent.DefaultPassword)
 	viper.SetDefault(flagRedact, dynamic.DefaultRedactFields)

types/entity.go

@@ -74,12 +74,11 @@ func GetEntitySubscription(entityID string) string {
 // FixtureEntity returns a testing fixture for an Entity object.
 func FixtureEntity(id string) *Entity {
 	return &Entity{
-		ID:               id,
-		Class:            "host",
-		Subscriptions:    []string{"linux"},
-		Environment:      "default",
-		Organization:     "default",
-		KeepaliveTimeout: 120,
+		ID:            id,
+		Class:         "host",
+		Subscriptions: []string{"linux"},
+		Environment:   "default",
+		Organization:  "default",
 	}
 }