feat(pkg/p2p): reconnect on disconnected peers

CHANGELOG.md

@@ -9,8 +9,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## v1.1.0-rc.2
+
 ### Changes
 
+- Improve P2P transient network failure [#3212](https://github.com/evstack/ev-node/pull/3212)
 * Improve execution/evm check for stored meta not stale [#3221](https://github.com/evstack/ev-node/pull/3221)
 
 ## v1.1.0-rc.1

apps/evm/go.mod

@@ -2,10 +2,10 @@ module github.com/evstack/ev-node/apps/evm
 
 go 1.25.7
 
-// replace (
-// 	github.com/evstack/ev-node => ../../
-// 	github.com/evstack/ev-node/execution/evm => ../../execution/evm
-// )
+replace (
+	github.com/evstack/ev-node => ../../
+	github.com/evstack/ev-node/execution/evm => ../../execution/evm
+)
 
 require (
 	github.com/ethereum/go-ethereum v1.17.2

apps/evm/go.sum

@@ -472,12 +472,8 @@ github.com/ethereum/go-bigmodexpfix v0.0.0-20250911101455-f9e208c548ab h1:rvv6MJ
 github.com/ethereum/go-bigmodexpfix v0.0.0-20250911101455-f9e208c548ab/go.mod h1:IuLm4IsPipXKF7CW5Lzf68PIbZ5yl7FFd74l/E0o9A8=
 github.com/ethereum/go-ethereum v1.17.2 h1:ag6geu0kn8Hv5FLKTpH+Hm2DHD+iuFtuqKxEuwUsDOI=
 github.com/ethereum/go-ethereum v1.17.2/go.mod h1:KHcRXfGOUfUmKg51IhQ0IowiqZ6PqZf08CMtk0g5K1o=
-github.com/evstack/ev-node v1.1.0-rc.1 h1:NtPuuDLqN2h4/edu5zxRlZAxmLkTG3ncXBO2PlCDvVs=
-github.com/evstack/ev-node v1.1.0-rc.1/go.mod h1:6rhWWzuyiqNn/erDmWCk1aLxUuQphyOGIRq56/smSyk=
 github.com/evstack/ev-node/core v1.0.0 h1:s0Tx0uWHme7SJn/ZNEtee4qNM8UO6PIxXnHhPbbKTz8=
 github.com/evstack/ev-node/core v1.0.0/go.mod h1:n2w/LhYQTPsi48m6lMj16YiIqsaQw6gxwjyJvR+B3sY=
-github.com/evstack/ev-node/execution/evm v1.0.0 h1:UTAdCrnPsLoGzSgsBx4Kv76jkXpMmHBIpNv3MxyzWPo=
-github.com/evstack/ev-node/execution/evm v1.0.0/go.mod h1:UrqkiepfTMiot6M8jnswgu3VU8SSucZpaMIHIl22/1A=
 github.com/fatih/color v1.7.0/go.mod h1:Zm6kSWBoL9eyXnKyktHP6abPY2pDugNf5KwzbycvMj4=
 github.com/fatih/color v1.10.0/go.mod h1:ELkj/draVOlAH/xkhN6mQ50Qd0MPOk5AAr3maGEBuJM=
 github.com/fatih/color v1.13.0/go.mod h1:kLAiJbzzSOZDVNGyDpeOxJ47H46qBXwg5ILebYFFOfk=

apps/testapp/go.mod

@@ -2,7 +2,7 @@ module github.com/evstack/ev-node/apps/testapp
 
 go 1.25.7
 
-// replace github.com/evstack/ev-node => ../../.
+replace github.com/evstack/ev-node => ../../.
 
 require (
 	github.com/evstack/ev-node v1.1.0-rc.1

apps/testapp/go.sum

@@ -432,8 +432,6 @@ github.com/envoyproxy/protoc-gen-validate v1.0.1/go.mod h1:0vj8bNkYbSTNS2PIyH87K
 github.com/envoyproxy/protoc-gen-validate v1.0.2/go.mod h1:GpiZQP3dDbg4JouG/NNS7QWXpgx6x8QiMKdmN72jogE=
 github.com/envoyproxy/protoc-gen-validate v1.3.0 h1:TvGH1wof4H33rezVKWSpqKz5NXWg5VPuZ0uONDT6eb4=
 github.com/envoyproxy/protoc-gen-validate v1.3.0/go.mod h1:HvYl7zwPa5mffgyeTUHA9zHIH36nmrm7oCbo4YKoSWA=
-github.com/evstack/ev-node v1.1.0-rc.1 h1:NtPuuDLqN2h4/edu5zxRlZAxmLkTG3ncXBO2PlCDvVs=
-github.com/evstack/ev-node v1.1.0-rc.1/go.mod h1:6rhWWzuyiqNn/erDmWCk1aLxUuQphyOGIRq56/smSyk=
 github.com/evstack/ev-node/core v1.0.0 h1:s0Tx0uWHme7SJn/ZNEtee4qNM8UO6PIxXnHhPbbKTz8=
 github.com/evstack/ev-node/core v1.0.0/go.mod h1:n2w/LhYQTPsi48m6lMj16YiIqsaQw6gxwjyJvR+B3sY=
 github.com/fatih/color v1.7.0/go.mod h1:Zm6kSWBoL9eyXnKyktHP6abPY2pDugNf5KwzbycvMj4=

pkg/p2p/client.go

@@ -35,6 +35,12 @@ const (
 
 	// peerLimit defines limit of number of peers returned during active peer discovery.
 	peerLimit = 60
+
+	// seedReconnectBackoff is the initial backoff when reconnecting to a disconnected seed peer.
+	seedReconnectBackoff = 1 * time.Second
+
+	// seedReconnectMaxBackoff is the maximum backoff for seed peer reconnection attempts.
+	seedReconnectMaxBackoff = 30 * time.Second
 )
 
 // Client is a P2P client, implemented with libp2p.
@@ -56,6 +62,11 @@ type Client struct {
 	ps      *pubsub.PubSub
 	started bool
 
+	ctx    context.Context
+	cancel context.CancelFunc
+
+	seedPeers []peer.AddrInfo
+
 	metrics *Metrics
 }
 
@@ -140,6 +151,7 @@ func (c *Client) Start(ctx context.Context) error {
 
 func (c *Client) startWithHost(ctx context.Context, h host.Host) error {
 	c.host = h
+	c.ctx, c.cancel = context.WithCancel(ctx)
 	for _, a := range c.host.Addrs() {
 		c.logger.Info().Str("address", fmt.Sprintf("%s/p2p/%s", a, c.host.ID())).Msg("listening on address")
 	}
@@ -170,11 +182,17 @@ func (c *Client) startWithHost(ctx context.Context, h host.Host) error {
 	}
 
 	c.started = true
+
+	c.host.Network().Notify(c.newDisconnectNotifee())
+
 	return nil
 }
 
 // Close gently stops Client.
 func (c *Client) Close() error {
+	if c.cancel != nil {
+		c.cancel()
+	}
 	var err error
 	if c.dht != nil {
 		err = errors.Join(err, c.dht.Close())
@@ -245,6 +263,77 @@ func (c *Client) Peers() []PeerConnection {
 	return res
 }
 
+type disconnectNotifee struct {
+	c *Client
+}
+
+func (n disconnectNotifee) Connected(_ network.Network, conn network.Conn) {
+	p := conn.RemotePeer()
+	for _, sp := range n.c.seedPeers {
+		if sp.ID == p {
+			n.c.logger.Info().Str("peer", p.String()).Msg("connected to seed peer")
+			return
+		}
+	}
+}
+func (n disconnectNotifee) OpenedStream(_ network.Network, _ network.Stream)     {}
+func (n disconnectNotifee) ClosedStream(_ network.Network, _ network.Stream)     {}
+func (n disconnectNotifee) Listen(_ network.Network, _ multiaddr.Multiaddr)      {}
+func (n disconnectNotifee) ListenClose(_ network.Network, _ multiaddr.Multiaddr) {}
+
+func (n disconnectNotifee) Disconnected(_ network.Network, conn network.Conn) {
+	p := conn.RemotePeer()
+	for _, sp := range n.c.seedPeers {
+		if sp.ID == p {
+			n.c.logger.Warn().Str("peer", p.String()).Msg("disconnected from seed peer, scheduling reconnect")
+			go n.c.reconnectSeedPeer(sp)
+			return
+		}
+	}
+}
+
+func (c *Client) reconnectSeedPeer(sp peer.AddrInfo) {
+	backoff := seedReconnectBackoff
+	for {
+		if c.ctx.Err() != nil {
+			return
+		}
+		if c.isConnected(sp.ID) {
+			return
+		}
+
+		err := c.host.Connect(c.ctx, sp)
+		if err == nil {
+			c.logger.Info().Str("peer", sp.ID.String()).Msg("reconnected to seed peer")
+			return
+		}
+		if c.ctx.Err() != nil {
+			return
+		}
+
+		c.logger.Debug().Str("peer", sp.ID.String()).Dur("backoff", backoff).Err(err).Msg("failed to reconnect to seed peer, retrying")
+		select {
+		case <-c.ctx.Done():
+			return
+		case <-time.After(backoff):
+		}
+
+		backoff *= 2
+		if backoff > seedReconnectMaxBackoff {
+			backoff = seedReconnectMaxBackoff
+		}
+	}
+}
+
+func (c *Client) newDisconnectNotifee() disconnectNotifee {
+	return disconnectNotifee{c: c}
+}
+
+// isConnected returns true if there is an active connection to the given peer.
+func (c *Client) isConnected(id peer.ID) bool {
+	return c.host.Network().Connectedness(id) == network.Connected
+}
+
 func (c *Client) listen() (host.Host, error) {
 	maddr, err := multiaddr.NewMultiaddr(c.conf.ListenAddress)
 	if err != nil {
@@ -256,6 +345,7 @@ func (c *Client) listen() (host.Host, error) {
 
 func (c *Client) setupDHT(ctx context.Context) error {
 	peers := c.parseAddrInfoList(c.conf.Peers)
+	c.seedPeers = peers
 	if len(peers) == 0 {
 		c.logger.Info().Msg("no peers - only listening for connections")
 	}

pkg/p2p/client_test.go

@@ -278,6 +278,104 @@ func waitForCondition(timeout time.Duration, conditionFunc func() bool) error {
 	}
 }
 
+func TestSeedPeerReconnect(t *testing.T) {
+	require := require.New(t)
+	assert := assert.New(t)
+	logger := zerolog.Nop()
+
+	mn := mocknet.New()
+	defer mn.Close()
+
+	seedKey, err := key.GenerateNodeKey()
+	require.NoError(err)
+	seedAddr, err := getAddr(seedKey.PrivKey)
+	require.NoError(err)
+	seedHost, err := mn.AddPeer(seedKey.PrivKey, seedAddr)
+	require.NoError(err)
+
+	clientKey, err := key.GenerateNodeKey()
+	require.NoError(err)
+	clientAddr, err := getAddr(clientKey.PrivKey)
+	require.NoError(err)
+	clientHost, err := mn.AddPeer(clientKey.PrivKey, clientAddr)
+	require.NoError(err)
+
+	seedAddrStr := seedHost.Addrs()[0].String() + "/p2p/" + seedHost.ID().String()
+	conf := config.P2PConfig{Peers: seedAddrStr}
+
+	client, err := NewClient(conf, clientKey.PrivKey, dssync.MutexWrap(datastore.NewMapDatastore()), "test-chain", logger, NopMetrics())
+	require.NoError(err)
+	require.NotNil(client)
+
+	err = mn.LinkAll()
+	require.NoError(err)
+	err = mn.ConnectAllButSelf()
+	require.NoError(err)
+
+	ctx := t.Context()
+	err = client.startWithHost(ctx, clientHost)
+	require.NoError(err)
+	defer client.Close()
+
+	err = waitForCondition(2*time.Second, func() bool {
+		return client.isConnected(seedHost.ID())
+	})
+	require.NoError(err, "client should connect to seed peer on start")
+
+	conns := client.host.Network().ConnsToPeer(seedHost.ID())
+	for _, conn := range conns {
+		conn.Close()
+	}
+	client.host.Network().ClosePeer(seedHost.ID())
+
+	assert.False(client.isConnected(seedHost.ID()), "seed peer should be disconnected")
+
+	err = waitForCondition(5*time.Second, func() bool {
+		return client.isConnected(seedHost.ID())
+	})
+	require.NoError(err, "client should reconnect to seed peer after disconnect")
+}
+
+func TestSeedPeerReconnectStopsOnClose(t *testing.T) {
+	require := require.New(t)
+
+	mn := mocknet.New()
+	defer mn.Close()
+
+	seedKey, err := key.GenerateNodeKey()
+	require.NoError(err)
+	seedAddr, err := getAddr(seedKey.PrivKey)
+	require.NoError(err)
+	seedHost, err := mn.AddPeer(seedKey.PrivKey, seedAddr)
+	require.NoError(err)
+
+	clientKey, err := key.GenerateNodeKey()
+	require.NoError(err)
+	clientAddr, err := getAddr(clientKey.PrivKey)
+	require.NoError(err)
+	clientHost, err := mn.AddPeer(clientKey.PrivKey, clientAddr)
+	require.NoError(err)
+
+	seedAddrStr := seedHost.Addrs()[0].String() + "/p2p/" + seedHost.ID().String()
+	conf := config.P2PConfig{Peers: seedAddrStr}
+
+	client, err := NewClient(conf, clientKey.PrivKey, dssync.MutexWrap(datastore.NewMapDatastore()), "test-chain", zerolog.Nop(), NopMetrics())
+	require.NoError(err)
+
+	err = mn.LinkAll()
+	require.NoError(err)
+	err = mn.ConnectAllButSelf()
+	require.NoError(err)
+
+	ctx := t.Context()
+	err = client.startWithHost(ctx, clientHost)
+	require.NoError(err)
+
+	require.NoError(client.Close())
+
+	require.Error(client.ctx.Err(), "client context should be cancelled after Close")
+}
+
 func TestClientInfoMethods(t *testing.T) {
 	require := require.New(t)
 	assert := assert.New(t)

pkg/signer/aws/signer.go

@@ -159,7 +159,7 @@ func (s *KmsSigner) Sign(ctx context.Context, message []byte) ([]byte, error) {
 	timeout := s.opts.timeout()
 	maxAttempts := maxRetries + 1
 
-	for attempt := 0; attempt < maxAttempts; attempt++ {
+	for attempt := range maxAttempts {
 		if err := ctx.Err(); err != nil {
 			return nil, err
 		}

pkg/signer/gcp/signer.go

@@ -189,7 +189,7 @@ func (s *KmsSigner) Sign(ctx context.Context, message []byte) ([]byte, error) {
 	timeout := s.opts.timeout()
 	maxAttempts := maxRetries + 1
 
-	for attempt := 0; attempt < maxAttempts; attempt++ {
+	for attempt := range maxAttempts {
 		if err := ctx.Err(); err != nil {
 			return nil, err
 		}