swarm: Fix T.Fatal inside a goroutine in tests (#18409)

* swarm/storage: fix T.Fatal inside a goroutine

* swarm/network/simulation: fix T.Fatal inside a goroutine

* swarm/network/stream: fix T.Fatal inside a goroutine

* swarm/network/simulation: consistent failures in TestPeerEventsTimeout

* swarm/network/simulation: rename sendRunSignal to triggerSimulationRun

4 files changed, 68 insertions, 17 deletions

diff --git a/swarm/network/stream/delivery_test.go b/swarm/network/stream/delivery_test.go
index 04a4bb077..c65e1386d 100644
--- a/swarm/network/stream/delivery_test.go
+++ b/swarm/network/stream/delivery_test.go
@@ -19,9 +19,11 @@ package stream
 import (
 	"bytes"
 	"context"
+	"errors"
 	"fmt"
 	"os"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -500,7 +502,7 @@ func testDeliveryFromNodes(t *testing.T, nodes, chunkCount int, skipCheck bool)
 
 	log.Info("Starting simulation")
 	ctx := context.Background()
-	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) error {
+	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) (err error) {
 		nodeIDs := sim.UpNodeIDs()
 		//determine the pivot node to be the first node of the simulation
 		pivot := nodeIDs[0]
@@ -553,14 +555,13 @@ func testDeliveryFromNodes(t *testing.T, nodes, chunkCount int, skipCheck bool)
 		}
 		pivotFileStore := item.(*storage.FileStore)
 		log.Debug("Starting retrieval routine")
+		retErrC := make(chan error)
 		go func() {
 			// start the retrieval on the pivot node - this will spawn retrieve requests for missing chunks
 			// we must wait for the peer connections to have started before requesting
 			n, err := readAll(pivotFileStore, fileHash)
 			log.Info(fmt.Sprintf("retrieved %v", fileHash), "read", n, "err", err)
-			if err != nil {
-				t.Fatalf("requesting chunks action error: %v", err)
-			}
+			retErrC <- err
 		}()
 
 		log.Debug("Watching for disconnections")
@@ -570,11 +571,19 @@ func testDeliveryFromNodes(t *testing.T, nodes, chunkCount int, skipCheck bool)
 			simulation.NewPeerEventsFilter().Drop(),
 		)
 
+		var disconnected atomic.Value
 		go func() {
 			for d := range disconnections {
 				if d.Error != nil {
 					log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
-					t.Fatal(d.Error)
+					disconnected.Store(true)
+				}
+			}
+		}()
+		defer func() {
+			if err != nil {
+				if yes, ok := disconnected.Load().(bool); ok && yes {
+					err = errors.New("disconnect events received")
 				}
 			}
 		}()
@@ -595,6 +604,9 @@ func testDeliveryFromNodes(t *testing.T, nodes, chunkCount int, skipCheck bool)
 		if !success {
 			return fmt.Errorf("Test failed, chunks not available on all nodes")
 		}
+		if err := <-retErrC; err != nil {
+			t.Fatalf("requesting chunks: %v", err)
+		}
 		log.Debug("Test terminated successfully")
 		return nil
 	})
@@ -675,7 +687,7 @@ func benchmarkDeliveryFromNodes(b *testing.B, nodes, chunkCount int, skipCheck b
 	}
 
 	ctx := context.Background()
-	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) error {
+	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) (err error) {
 		nodeIDs := sim.UpNodeIDs()
 		node := nodeIDs[len(nodeIDs)-1]
 
@@ -702,11 +714,19 @@ func benchmarkDeliveryFromNodes(b *testing.B, nodes, chunkCount int, skipCheck b
 			simulation.NewPeerEventsFilter().Drop(),
 		)
 
+		var disconnected atomic.Value
 		go func() {
 			for d := range disconnections {
 				if d.Error != nil {
 					log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
-					b.Fatal(d.Error)
+					disconnected.Store(true)
+				}
+			}
+		}()
+		defer func() {
+			if err != nil {
+				if yes, ok := disconnected.Load().(bool); ok && yes {
+					err = errors.New("disconnect events received")
 				}
 			}
 		}()
diff --git a/swarm/network/stream/intervals_test.go b/swarm/network/stream/intervals_test.go
index 7c7feeb11..121758c18 100644
--- a/swarm/network/stream/intervals_test.go
+++ b/swarm/network/stream/intervals_test.go
@@ -19,9 +19,11 @@ package stream
 import (
 	"context"
 	"encoding/binary"
+	"errors"
 	"fmt"
 	"os"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -117,7 +119,7 @@ func testIntervals(t *testing.T, live bool, history *Range, skipCheck bool) {
 		t.Fatal(err)
 	}
 
-	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) error {
+	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) (err error) {
 		nodeIDs := sim.UpNodeIDs()
 		storer := nodeIDs[0]
 		checker := nodeIDs[1]
@@ -162,11 +164,19 @@ func testIntervals(t *testing.T, live bool, history *Range, skipCheck bool) {
 			return err
 		}
 
+		var disconnected atomic.Value
 		go func() {
 			for d := range disconnections {
 				if d.Error != nil {
 					log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
-					t.Fatal(d.Error)
+					disconnected.Store(true)
+				}
+			}
+		}()
+		defer func() {
+			if err != nil {
+				if yes, ok := disconnected.Load().(bool); ok && yes {
+					err = errors.New("disconnect events received")
 				}
 			}
 		}()
diff --git a/swarm/network/stream/snapshot_sync_test.go b/swarm/network/stream/snapshot_sync_test.go
index f86d9acca..7a883644b 100644
--- a/swarm/network/stream/snapshot_sync_test.go
+++ b/swarm/network/stream/snapshot_sync_test.go
@@ -21,6 +21,7 @@ import (
 	"os"
 	"runtime"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -213,11 +214,13 @@ func testSyncingViaGlobalSync(t *testing.T, chunkCount int, nodeCount int) {
 		simulation.NewPeerEventsFilter().Drop(),
 	)
 
+	var disconnected atomic.Value
 	go func() {
 		for d := range disconnections {
-			log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
-			t.Fatal("unexpected disconnect")
-			cancelSimRun()
+			if d.Error != nil {
+				log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
+				disconnected.Store(true)
+			}
 		}
 	}()
 
@@ -226,6 +229,9 @@ func testSyncingViaGlobalSync(t *testing.T, chunkCount int, nodeCount int) {
 	if result.Error != nil {
 		t.Fatal(result.Error)
 	}
+	if yes, ok := disconnected.Load().(bool); ok && yes {
+		t.Fatal("disconnect events received")
+	}
 	log.Info("Simulation ended")
 }
 
@@ -395,11 +401,13 @@ func testSyncingViaDirectSubscribe(t *testing.T, chunkCount int, nodeCount int)
 		simulation.NewPeerEventsFilter().Drop(),
 	)
 
+	var disconnected atomic.Value
 	go func() {
 		for d := range disconnections {
-			log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
-			t.Fatal("unexpected disconnect")
-			cancelSimRun()
+			if d.Error != nil {
+				log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
+				disconnected.Store(true)
+			}
 		}
 	}()
 
@@ -514,6 +522,9 @@ func testSyncingViaDirectSubscribe(t *testing.T, chunkCount int, nodeCount int)
 		return result.Error
 	}
 
+	if yes, ok := disconnected.Load().(bool); ok && yes {
+		t.Fatal("disconnect events received")
+	}
 	log.Info("Simulation ended")
 	return nil
 }
diff --git a/swarm/network/stream/syncer_test.go b/swarm/network/stream/syncer_test.go
index 2fdba71ff..09152ebfe 100644
--- a/swarm/network/stream/syncer_test.go
+++ b/swarm/network/stream/syncer_test.go
@@ -18,11 +18,13 @@ package stream
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io/ioutil"
 	"math"
 	"os"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -129,7 +131,7 @@ func testSyncBetweenNodes(t *testing.T, nodes, chunkCount int, skipCheck bool, p
 	if err != nil {
 		t.Fatal(err)
 	}
-	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) error {
+	result := sim.Run(ctx, func(ctx context.Context, sim *simulation.Simulation) (err error) {
 		nodeIDs := sim.UpNodeIDs()
 
 		nodeIndex := make(map[enode.ID]int)
@@ -143,11 +145,19 @@ func testSyncBetweenNodes(t *testing.T, nodes, chunkCount int, skipCheck bool, p
 			simulation.NewPeerEventsFilter().Drop(),
 		)
 
+		var disconnected atomic.Value
 		go func() {
 			for d := range disconnections {
 				if d.Error != nil {
 					log.Error("peer drop", "node", d.NodeID, "peer", d.PeerID)
-					t.Fatal(d.Error)
+					disconnected.Store(true)
+				}
+			}
+		}()
+		defer func() {
+			if err != nil {
+				if yes, ok := disconnected.Load().(bool); ok && yes {
+					err = errors.New("disconnect events received")
 				}
 			}
 		}()