Suppresses heartbeat related start-actual-lrp log messages to save log volume

Adds configuration to allow this feature to be enabled/disabled

Author: geofffranks

cmd/bbs/config/config.go

@@ -38,6 +38,7 @@ type BBSConfig struct {
 	ConvergenceWorkers            int                   `json:"convergence_workers,omitempty"`
 	DatabaseConnectionString      string                `json:"database_connection_string"`
 	DatabaseDriver                string                `json:"database_driver,omitempty"`
+	DebugLRPStartHeartbeats       bool                  `json:"debug_lrp_start_heartbeats,omitempty"`
 	DesiredLRPCreationTimeout     durationjson.Duration `json:"desired_lrp_creation_timeout,omitempty"`
 	ExpireCompletedTaskDuration   durationjson.Duration `json:"expire_completed_task_duration,omitempty"`
 	ExpirePendingTaskDuration     durationjson.Duration `json:"expire_pending_task_duration,omitempty"`

cmd/bbs/config/config_test.go

@@ -40,6 +40,7 @@ var _ = Describe("BBSConfig", func() {
 			"database_connection_string": "",
 			"database_driver": "postgres",
 			"debug_address": "127.0.0.1:17017",
+			"debug_lrp_start_heartbeats":true,
 			"desired_lrp_creation_timeout": "1m0s",
 			"encryption_keys": {"label": "key"},
 			"expire_completed_task_duration": "2m0s",
@@ -137,6 +138,7 @@ var _ = Describe("BBSConfig", func() {
 			DebugServerConfig: debugserver.DebugServerConfig{
 				DebugAddress: "127.0.0.1:17017",
 			},
+			DebugLRPStartHeartbeats:   true,
 			DesiredLRPCreationTimeout: durationjson.Duration(1 * time.Minute),
 			EncryptionConfig: encryption.EncryptionConfig{
 				ActiveKeyLabel: "label",

cmd/bbs/lrp_convergence_test.go

@@ -111,6 +111,7 @@ var _ = Describe("Convergence API", func() {
 						clock.NewClock(),
 						sqlRunner.DriverName(),
 						metronClient,
+						false,
 					)
 
 					Eventually(func() models.ActualLRP_Presence {
@@ -176,6 +177,7 @@ var _ = Describe("Convergence API", func() {
 							map[string]string{},
 							false,
 							"",
+							false,
 						)
 						Expect(err).NotTo(HaveOccurred())
 					})

cmd/bbs/main.go

@@ -137,6 +137,7 @@ func main() {
 		clock,
 		bbsConfig.DatabaseDriver,
 		metronClient,
+		bbsConfig.DebugLRPStartHeartbeats,
 	)
 	err = sqlDB.CreateConfigurationsTable(context.Background(), logger)
 	if err != nil {

controllers/actual_lrp_lifecycle_controller.go

@@ -132,9 +132,11 @@ func (h *ActualLRPLifecycleController) StartActualLRP(ctx context.Context,
 		return nil
 	}
 
+	isCurrentlyRunning := lrp != nil && lrp.State == models.ActualLRPStateRunning
+
 	// creates ordinary running actual LRP if it doesn't exist, otherwise updates
 	// the existing ordinary actual LRP to running state
-	before, after, err := h.db.StartActualLRP(ctx, logger, actualLRPKey, actualLRPInstanceKey, actualLRPNetInfo, actualLRPInternalRoutes, actualLRPMetricTags, routable, availabilityZone)
+	before, after, err := h.db.StartActualLRP(ctx, logger, actualLRPKey, actualLRPInstanceKey, actualLRPNetInfo, actualLRPInternalRoutes, actualLRPMetricTags, routable, availabilityZone, isCurrentlyRunning)
 	if err != nil {
 		return err
 	}

controllers/actual_lrp_lifecycle_controller_test.go

@@ -421,7 +421,7 @@ var _ = Describe("ActualLRP Lifecycle Controller", func() {
 				Expect(fakeActualLRPDB.StartActualLRPCallCount()).To(Equal(1))
 				Expect(fakeActualLRPDB.ActualLRPsCallCount()).To(Equal(1))
 
-				_, _, _, _, _, internalRoutesArgument, metricTagsArgument, routableArgument, availabilityZoneArgument := fakeActualLRPDB.StartActualLRPArgsForCall(0)
+				_, _, _, _, _, internalRoutesArgument, metricTagsArgument, routableArgument, availabilityZoneArgument, _ := fakeActualLRPDB.StartActualLRPArgsForCall(0)
 
 				Expect(err).NotTo(HaveOccurred())
 				Expect(internalRoutesArgument).To(Equal(internalRoutes))
@@ -532,11 +532,23 @@ var _ = Describe("ActualLRP Lifecycle Controller", func() {
 					createdEvent := event.(*models.ActualLRPInstanceCreatedEvent)
 					Expect(createdEvent.ActualLrp).To(Equal(afterActualLRP))
 				})
+				It("tells the db that a previous instance didn't exist", func() {
+					err = controller.StartActualLRP(ctx, logger, &actualLRPKey, &afterInstanceKey, &netInfo, internalRoutes, metricTags, routable, availabilityZone)
+					Expect(err).NotTo(HaveOccurred())
+
+					Expect(fakeActualLRPDB.StartActualLRPCallCount()).To(Equal(1))
+					_, _, _, _, _, _, _, _, _, isRunning := fakeActualLRPDB.StartActualLRPArgsForCall(0)
+					Expect(isRunning).To(BeFalse())
+				})
 			})
 
 			Context("when the actual lrp was updated", func() {
+				JustBeforeEach(func() {
+					actualLRP.State = models.ActualLRPStateRunning
+					fakeActualLRPDB.ActualLRPsReturns([]*models.ActualLRP{actualLRP}, nil)
+				})
 				It("emits a change event to the hub", func() {
-					err = controller.StartActualLRP(ctx, logger, &actualLRPKey, &afterInstanceKey, &netInfo, internalRoutes, metricTags, routable, availabilityZone)
+					err = controller.StartActualLRP(ctx, logger, &actualLRPKey, &beforeInstanceKey, &netInfo, internalRoutes, metricTags, routable, availabilityZone)
 					Eventually(actualHub.EmitCallCount).Should(Equal(1))
 					event := actualHub.EmitArgsForCall(0)
 					//lint:ignore SA1019 - calling deprecated model while unit testing deprecated method
@@ -547,6 +559,14 @@ var _ = Describe("ActualLRP Lifecycle Controller", func() {
 					//lint:ignore SA1019 - still need to emit these events until the ActaulLRPGroup api is deleted
 					Expect(changedEvent.After).To(Equal(afterActualLRP.ToActualLRPGroup()))
 				})
+				It("tells the db that a previous instance exists", func() {
+					err = controller.StartActualLRP(ctx, logger, &actualLRPKey, &afterInstanceKey, &netInfo, internalRoutes, metricTags, routable, availabilityZone)
+					Expect(err).NotTo(HaveOccurred())
+
+					Expect(fakeActualLRPDB.StartActualLRPCallCount()).To(Equal(1))
+					_, _, _, _, _, _, _, _, _, isRunning := fakeActualLRPDB.StartActualLRPArgsForCall(0)
+					Expect(isRunning).To(BeTrue())
+				})
 			})
 
 			Context("when Routable was updated", func() {

db/actual_lrp_db.go

@@ -24,6 +24,7 @@ type ActualLRPDB interface {
 		metricTags map[string]string,
 		routable bool,
 		availabilityZone string,
+		isCurrentlyRunning bool,
 	) (before *models.ActualLRP, after *models.ActualLRP, err error)
 	CrashActualLRP(ctx context.Context, logger lager.Logger, key *models.ActualLRPKey, instanceKey *models.ActualLRPInstanceKey, crashReason string) (before *models.ActualLRP, after *models.ActualLRP, shouldRestart bool, err error)
 	FailActualLRP(ctx context.Context, logger lager.Logger, key *models.ActualLRPKey, placementError string) (before *models.ActualLRP, after *models.ActualLRP, err error)