fix(ci): Fix unit flakes from testnode by stopping calling the stop functions multiple times

core/exchange_test.go

@@ -186,8 +186,13 @@ func TestExchange_StoreHistoricIfArchival(t *testing.T) {
 }
 
 func createCoreFetcher(t *testing.T, cfg *testnode.Config) (*BlockFetcher, *Network) {
+	t.Helper()
+
 	network := NewNetwork(t, cfg)
 	require.NoError(t, network.Start())
+	t.Cleanup(func() {
+		require.NoError(t, network.Stop())
+	})
 	// wait for height 2 in order to be able to start submitting txs (this prevents
 	// flakiness with accessing account state)
 	_, err := network.WaitForHeightWithTimeout(2, time.Second*2) // TODO @renaynay: configure?

core/testing.go

@@ -2,8 +2,10 @@ package core
 
 import (
 	"context"
+	"errors"
 	"net"
 	"path/filepath"
+	"sync"
 	"testing"
 	"time"
 
@@ -117,6 +119,9 @@ type Network struct {
 	stopNode func() error
 	stopGRPC func() error
 	stopAPI  func() error
+
+	stopOnce sync.Once
+	stopErr  error
 }
 
 func NewNetwork(t testing.TB, config *testnode.Config) *Network {
@@ -182,19 +187,29 @@ func (n *Network) Start() error {
 }
 
 func (n *Network) Stop() error {
-	err := n.stopNode()
-	if err != nil {
-		return err
-	}
-
-	err = n.stopGRPC()
-	if err != nil {
-		return err
-	}
+	n.stopOnce.Do(func() {
+		var errs []error
+
+		if n.stopNode != nil {
+			if err := n.stopNode(); err != nil {
+				errs = append(errs, err)
+			}
+		}
+
+		if n.stopGRPC != nil {
+			if err := n.stopGRPC(); err != nil {
+				errs = append(errs, err)
+			}
+		}
+
+		if n.stopAPI != nil {
+			if err := n.stopAPI(); err != nil {
+				errs = append(errs, err)
+			}
+		}
+
+		n.stopErr = errors.Join(errs...)
+	})
 
-	err = n.stopAPI()
-	if err != nil {
-		return err
-	}
-	return nil
+	return n.stopErr
 }

nodebuilder/tests/helpers_test.go

@@ -26,8 +26,9 @@ func getAdminClient(ctx context.Context, nd *nodebuilder.Node, t *testing.T) *cl
 	jwt, err := authtoken.NewSignedJWT(signer, []auth.Permission{"public", "read", "write", "admin"}, time.Minute)
 	require.NoError(t, err)
 
-	client, err := client.NewClient(ctx, listenAddr, jwt)
+	client, err := client.NewClient(context.WithoutCancel(ctx), listenAddr, jwt)
 	require.NoError(t, err)
+	t.Cleanup(client.Close)
 
 	return client
 }

nodebuilder/tests/prune_test.go

@@ -18,7 +18,6 @@ import (
 
 	"github.com/celestiaorg/celestia-node/blob"
 	"github.com/celestiaorg/celestia-node/nodebuilder"
-	"github.com/celestiaorg/celestia-node/nodebuilder/das"
 	"github.com/celestiaorg/celestia-node/nodebuilder/node"
 	"github.com/celestiaorg/celestia-node/nodebuilder/pruner"
 	"github.com/celestiaorg/celestia-node/nodebuilder/tests/swamp"
@@ -100,11 +99,10 @@ func TestArchivalBlobSync(t *testing.T) {
 	err = pruningBN.Start(ctx)
 	require.NoError(t, err)
 
-	pruningCfg := nodebuilder.DefaultConfig(node.Bridge)
-	pruningCfg.DASer = das.DefaultConfig(node.Full)
 	pruningFulls := make([]*nodebuilder.Node, 0, 3)
 	for range 3 {
-		pruningFN := sw.NewNodeWithConfig(node.Full, pruningCfg, prunerOpts)
+		cfg := sw.DefaultTestConfig(node.Full)
+		pruningFN := sw.NewNodeWithConfig(node.Full, cfg, prunerOpts)
 		err = pruningFN.Start(ctx)
 		require.NoError(t, err)
 
@@ -186,15 +184,14 @@ func TestDisallowConvertFromPrunedToArchival(t *testing.T) {
 
 	// Light nodes have pruning enabled by default
 	for _, nt := range []node.Type{node.Bridge, node.Full} {
-		store := nodebuilder.MockStore(t, nodebuilder.DefaultConfig(nt))
+		store := nodebuilder.MockStore(t, sw.DefaultTestConfig(nt))
 		pruningNode := sw.MustNewNodeWithStore(nt, store)
 		err := pruningNode.Start(ctx)
 		require.NoError(t, err)
 		err = pruningNode.Stop(ctx)
 		require.NoError(t, err)
 
-		archivalCfg := sw.DefaultTestConfig(nt)
-		err = store.PutConfig(archivalCfg)
+		err = store.PutConfig(sw.DefaultTestConfig(nt))
 		require.NoError(t, err)
 		// fx.Replace simulates the `--archival` flag being passed
 		pruningNode, err = sw.NewNodeWithStore(nt, store, fx.Replace(&pruner.Config{EnableService: false}))
@@ -222,7 +219,7 @@ func TestDisallowConvertToArchivalViaLastPrunedCheck(t *testing.T) {
 	}
 
 	for _, nt := range []node.Type{node.Bridge, node.Full} {
-		store := nodebuilder.MockStore(t, nodebuilder.DefaultConfig(nt))
+		store := nodebuilder.MockStore(t, sw.DefaultTestConfig(nt))
 		ds, err := store.Datastore()
 		require.NoError(t, err)
 
@@ -290,7 +287,7 @@ func TestConvertFromArchivalToPruned(t *testing.T) {
 		require.NoError(t, err)
 
 		// convert to pruned node
-		err = store.PutConfig(nodebuilder.DefaultConfig(nt))
+		err = store.PutConfig(sw.DefaultTestConfig(nt))
 		require.NoError(t, err)
 		prunedNd, err := sw.NewNodeWithStore(nt, store)
 		require.NoError(t, err)

share/eds/retriever_test.go

@@ -25,7 +25,7 @@ import (
 func TestRetriever_Retrieve(t *testing.T) {
 	// TODO @node-team: figure out why this regressed in CI
 	t.Skip("skipping retrieval as dangling component")
-	ctx, cancel := context.WithCancel(context.Background())
+	baseCtx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
 	bServ := ipld.NewMemBlockservice()
@@ -50,13 +50,11 @@ func TestRetriever_Retrieve(t *testing.T) {
 			// generate EDS
 			shares, err := libshare.RandShares(tc.squareSize * tc.squareSize)
 			require.NoError(t, err)
+			ctx, cancel := context.WithTimeout(baseCtx, time.Minute*5) // generous timeout for large squares
+			t.Cleanup(cancel)
 			in, err := ipld.AddShares(ctx, shares, bServ)
 			require.NoError(t, err)
 
-			// limit with timeout, specifically retrieval
-			ctx, cancel := context.WithTimeout(ctx, time.Minute*5) // the timeout is big for the max size which is long
-			defer cancel()
-
 			roots, err := share.NewAxisRoots(in)
 			require.NoError(t, err)
 			out, err := r.Retrieve(ctx, roots)
@@ -99,13 +97,14 @@ func TestRetriever_MultipleRandQuadrants(t *testing.T) {
 }
 
 func TestFraudProofValidation(t *testing.T) {
-	ctx, cancel := context.WithTimeout(context.Background(), time.Second*20)
-	defer t.Cleanup(cancel)
 	bServ := ipld.NewMemBlockservice()
 
 	odsSize := []int{2, 4, 16, 32, 64, 128}
 	for _, size := range odsSize {
 		t.Run(fmt.Sprintf("ods size:%d", size), func(t *testing.T) {
+			ctx, cancel := context.WithTimeout(context.Background(), time.Minute*5)
+			t.Cleanup(cancel)
+
 			var errByz *byzantine.ErrByzantine
 			faultHeader, err := generateByzantineError(ctx, t, size, bServ)
 			require.NotNil(t, err)