Add configurable mediorum repair interval and enable/disable

pkg/mediorum/mediorum.go

@@ -136,6 +136,17 @@ func runMediorum(lc *lifecycle.Lifecycle, logger *zap.Logger, mediorumEnv string
 		moveFromBlobStoreDSN = getenvWithDefault("OPENAUDIO_STORAGE_DRIVER_URL_MOVE_FROM", os.Getenv("AUDIUS_STORAGE_DRIVER_URL_MOVE_FROM"))
 	}
 
+	// Repair configuration
+	repairEnabled := getenvWithDefault("OPENAUDIO_REPAIR_ENABLED", "true") == "true"
+	repairInterval := time.Hour
+	if ri := os.Getenv("OPENAUDIO_REPAIR_INTERVAL"); ri != "" {
+		if parsed, err := time.ParseDuration(ri); err == nil {
+			repairInterval = parsed
+		} else {
+			logger.Warn("failed to parse OPENAUDIO_REPAIR_INTERVAL, using default 1h", zap.String("value", ri), zap.Error(err))
+		}
+	}
+
 	config := server.MediorumConfig{
 		Self: registrar.Peer{
 			Host:   httputil.RemoveTrailingSlash(strings.ToLower(nodeEndpoint)),
@@ -161,6 +172,8 @@ func runMediorum(lc *lifecycle.Lifecycle, logger *zap.Logger, mediorumEnv string
 		DiscoveryListensEndpoints: discoveryListensEndpoints(),
 		LogLevel:                  getenvWithDefault("OPENAUDIO_LOG_LEVEL", "info"),
 		DeadHosts:                 []string{},
+		RepairEnabled:             repairEnabled,
+		RepairInterval:            repairInterval,
 	}
 
 	ss, err := server.New(lc, logger, config, posChannel, core, ethService)

pkg/mediorum/server/repair.go

@@ -41,13 +41,21 @@ type RepairTracker struct {
 func (ss *MediorumServer) startRepairer(ctx context.Context) error {
 	logger := ss.logger.With(zap.String("task", "repair"))
 
+	if !ss.Config.RepairEnabled {
+		logger.Info("repair is disabled via OPENAUDIO_REPAIR_ENABLED=false")
+		<-ctx.Done()
+		return ctx.Err()
+	}
+
+	repairInterval := ss.Config.RepairInterval
+	logger.Info("repair configured", zap.Duration("interval", repairInterval))
+
 	// wait a minute on startup to determine healthy peers
 	ticker := time.NewTicker(1 * time.Minute)
 	for {
 		select {
 		case <-ticker.C:
-			// Wait 1 hour for next interval unless otherwise specified
-			ticker.Reset(1 * time.Hour)
+			ticker.Reset(repairInterval)
 
 			// pick up where we left off from the last repair.go run, including if the server restarted in the middle of a run
 			tracker := RepairTracker{

pkg/mediorum/server/server.go

@@ -75,6 +75,8 @@ type MediorumConfig struct {
 	DiscoveryListensEndpoints []string
 	LogLevel                  string
 	DeadHosts                 []string
+	RepairEnabled             bool          `default:"true"`
+	RepairInterval            time.Duration `default:"1h"`
 
 	ProgrammableDistributionEnabled bool