chore: fix panic from dependency creation race (#2233)

This change fixes a bug where restored experiments would attempt to send an ExperimentCreated message to the hp importance actor prior to its creation, resulting in a panic from a send on a nil actor.

master/internal/core.go

@@ -585,6 +585,12 @@ func (m *Master) Run(ctx context.Context) error {
 
 	m.system.MustActorOf(actor.Addr("allocation-aggregator"), &allocationAggregator{db: m.db})
 
+	hpi, err := hpimportance.NewManager(m.db, m.system, m.config.HPImportance, m.config.Root)
+	if err != nil {
+		return err
+	}
+	m.hpImportance, _ = m.system.ActorOf(actor.Addr(hpimportance.RootAddr), hpi)
+
 	// Initialize the HTTP server and listen for incoming requests.
 	m.echo = echo.New()
 	m.echo.Use(middleware.Recover())
@@ -791,11 +797,5 @@ func (m *Master) Run(ctx context.Context) error {
 		log.Info("telemetry reporting is disabled")
 	}
 
-	hpi, err := hpimportance.NewManager(m.db, m.system, m.config.HPImportance, m.config.Root)
-	if err != nil {
-		return err
-	}
-	m.hpImportance, _ = m.system.ActorOf(actor.Addr(hpimportance.RootAddr), hpi)
-
 	return m.startServers(ctx, cert)
 }

master/internal/experiment.go

@@ -191,8 +191,8 @@ func (e *experiment) Receive(ctx *actor.Context) error {
 			e.restoreTrialsFromPriorOperations(ctx, e.searcher.TrialOperations)
 		} else {
 			e.processOperations(ctx, e.searcher.TrialOperations, nil)
+			ctx.Tell(e.hpImportance, hpimportance.ExperimentCreated{ID: e.ID})
 		}
-		ctx.Tell(e.hpImportance, hpimportance.ExperimentCreated{ID: e.ID})
 		// Since e.searcher.TrialOperations should have all trials that were previously
 		// allocated, we can stop trying to restore new trials after processing these.
 		e.restored = false