refactor(monitoring): allow disabling of certain healthz checks over config

Author: invis-bitfly

backend/pkg/api/data_access/healthz.go

@@ -10,6 +10,7 @@ import (
 	"github.com/gobitfly/beaconchain/pkg/commons/log"
 	"github.com/gobitfly/beaconchain/pkg/commons/utils"
 	"github.com/gobitfly/beaconchain/pkg/monitoring/constants"
+	monitoringServices "github.com/gobitfly/beaconchain/pkg/monitoring/services"
 )
 
 type HealthzRepository interface {
@@ -128,10 +129,7 @@ func (d *DataAccessService) GetHealthz(ctx context.Context, showAll bool) types.
 	for _, result := range results {
 		response.Reports[result.EventId] = append(response.Reports[result.EventId], result)
 	}
-	requiredEvents := constants.RequiredEvents
-	if utils.Config.DeploymentType == "production" {
-		requiredEvents = append(requiredEvents, constants.ProductionRequiredEvents...)
-	}
+	requiredEvents := monitoringServices.GetRequiredEvents()
 	for _, id := range requiredEvents {
 		if _, ok := response.Reports[string(id)]; !ok {
 			response.Reports[string(id)] = []types.HealthzResult{

backend/pkg/exporter/modules/pubkey_tags.go

@@ -12,9 +12,10 @@ import (
 
 func UpdatePubkeyTag() {
 	log.Infof("Started Pubkey Tags Updater")
+	delay := time.Minute * 10
 	for {
 		start := time.Now()
-		r := monitoringServices.NewStatusReport(constants.Event_ExporterLegacyPubkeyTags, constants.Default, time.Second*12)
+		r := monitoringServices.NewStatusReport(constants.Event_ExporterLegacyPubkeyTags, delay, time.Second*12)
 		r(constants.Running, nil)
 		tx, err := db.WriterDb.Beginx()
 		if err != nil {
@@ -45,6 +46,6 @@ func UpdatePubkeyTag() {
 		r(constants.Success, map[string]string{"took": time.Since(start).String(), "took_raw": time.Since(start).String()})
 		metrics.TaskDuration.WithLabelValues("validator_pubkey_tag_updater").Observe(time.Since(start).Seconds())
 
-		time.Sleep(time.Minute * 10)
+		time.Sleep(delay)
 	}
 }

backend/pkg/exporter/modules/rocketpool.go

@@ -191,7 +191,7 @@ func (rp *RocketpoolExporter) Run() error {
 
 	for {
 		t0 := time.Now()
-		r := monitoringServices.NewStatusReport(constants.Event_ExporterLegacyRocketPool, constants.Default, time.Second*12)
+		r := monitoringServices.NewStatusReport(constants.Event_ExporterLegacyRocketPool, time.Hour*4, rp.UpdateInterval) // currently takes 2h40m on mainnet...
 		r(constants.Running, nil)
 		var err error
 		err = rp.Update(count)

backend/pkg/monitoring/constants/main.go

@@ -1,6 +1,8 @@
 package constants
 
-import "time"
+import (
+	"time"
+)
 
 // status enum
 type StatusType string
@@ -69,8 +71,6 @@ var ProductionRequiredEvents = []Event{
 	Event_ExporterLegacyNetworkLiveness,
 	Event_ExporterLegacySyncCommittees,
 	Event_ExporterLegacySyncCommitteesCount,
-	Event_ExporterLegacyRocketPool,
-	Event_ExporterLegacyPubkeyTags,
 	Event_ExporterModuleELRewardsFinalizer,
 	Event_ExporterModuleELPayloadExporter,
 	Event_ExporterModuleELDepositsExporter,

backend/pkg/monitoring/services/base.go

@@ -109,3 +109,18 @@ func NewStatusReport(id constants.Event, timeout time.Duration, check_interval t
 		}()
 	}
 }
+
+func GetRequiredEvents() []constants.Event {
+	// i would hope this simple of a function doesnt need caching
+	requiredEvents := constants.RequiredEvents
+	if utils.Config.DeploymentType == "production" {
+		requiredEvents = append(requiredEvents, constants.ProductionRequiredEvents...)
+	}
+	if utils.Config.RocketpoolExporter.Enabled {
+		requiredEvents = append(requiredEvents, constants.Event_ExporterLegacyRocketPool)
+	}
+	if utils.Config.Indexer.PubKeyTagsExporter.Enabled {
+		requiredEvents = append(requiredEvents, constants.Event_ExporterLegacyPubkeyTags)
+	}
+	return requiredEvents
+}