Backport release/v6.6: fix(p2p): make the inbound accept rate configurable and raise its default

sei-tendermint/config/config.go

@@ -140,6 +140,9 @@ func (cfg *Config) ValidateBasic() error {
 	if err := cfg.RPC.ValidateBasic(); err != nil {
 		return fmt.Errorf("error in [rpc] section: %w", err)
 	}
+	if err := cfg.P2P.ValidateBasic(); err != nil {
+		return fmt.Errorf("error in [p2p] section: %w", err)
+	}
 	if err := cfg.Mempool.ValidateBasic(); err != nil {
 		return fmt.Errorf("error in [mempool] section: %w", err)
 	}
@@ -711,6 +714,10 @@ type P2PConfig struct {
 	// How often node should dial a new peer.
 	DialInterval time.Duration `mapstructure:"dial-interval"`
 
+	// How often node should accept a new inbound connection. A value of 0 disables
+	// the limiter.
+	AcceptInterval time.Duration `mapstructure:"accept-interval"`
+
 	// Testing params.
 	// Force dial to fail
 	TestDialFail bool `mapstructure:"test-dial-fail"`
@@ -741,6 +748,7 @@ func DefaultP2PConfig() *P2PConfig {
 		HandshakeTimeout:              10 * time.Second,
 		DialTimeout:                   3 * time.Second,
 		DialInterval:                  10 * time.Second,
+		AcceptInterval:                10 * time.Millisecond,
 		TestDialFail:                  false,
 		QueueType:                     "simple-priority",
 	}
@@ -761,6 +769,12 @@ func (cfg *P2PConfig) ValidateBasic() error {
 	if cfg.RecvRate < 0 {
 		return errors.New("recv-rate can't be negative")
 	}
+	if cfg.DialInterval < 0 {
+		return errors.New("dial-interval can't be negative")
+	}
+	if cfg.AcceptInterval < 0 {
+		return errors.New("accept-interval can't be negative")
+	}
 	return nil
 }
 

sei-tendermint/config/config_test.go

@@ -11,6 +11,7 @@ import (
 	"github.com/sei-protocol/sei-chain/sei-tendermint/types"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"golang.org/x/time/rate"
 )
 
 func TestDefaultConfig(t *testing.T) {
@@ -38,6 +39,16 @@ func TestConfigValidateBasic(t *testing.T) {
 	assert.Error(t, cfg.ValidateBasic())
 }
 
+// Asserts Config.ValidateBasic routes the [p2p] section, not merely that the
+// section's own checks work.
+func TestConfigValidateBasicRoutesP2P(t *testing.T) {
+	cfg := DefaultConfig()
+	require.NoError(t, cfg.ValidateBasic())
+
+	cfg.P2P.AcceptInterval = -1
+	require.Error(t, cfg.ValidateBasic())
+}
+
 func TestTLSConfiguration(t *testing.T) {
 	cfg := DefaultConfig()
 	cfg.SetRoot("/home/user")
@@ -229,6 +240,8 @@ func TestP2PConfigValidateBasic(t *testing.T) {
 		"MaxPacketMsgPayloadSize",
 		"SendRate",
 		"RecvRate",
+		"DialInterval",
+		"AcceptInterval",
 	}
 
 	for _, fieldName := range fieldsToTest {
@@ -238,6 +251,21 @@ func TestP2PConfigValidateBasic(t *testing.T) {
 	}
 }
 
+// Pins the accept-interval default exactly, so changing it is deliberate and
+// visible in the diff.
+func TestP2PConfigAcceptInterval(t *testing.T) {
+	cfg := DefaultP2PConfig()
+	require.NoError(t, cfg.ValidateBasic())
+
+	require.Equal(t, 10*time.Millisecond, cfg.AcceptInterval)
+
+	// A zero interval is the documented escape hatch for disabling the limiter
+	// outright, and must stay valid rather than becoming a zero rate.
+	cfg.AcceptInterval = 0
+	require.NoError(t, cfg.ValidateBasic())
+	require.Equal(t, rate.Inf, rate.Every(cfg.AcceptInterval))
+}
+
 // --- WalFile legacy fallback tests ---
 
 func TestWalFile_NewDefault_NoLegacy(t *testing.T) {

sei-tendermint/config/p2p_compat_test.go

@@ -0,0 +1,61 @@
+package config_test
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/spf13/viper"
+	"github.com/stretchr/testify/require"
+
+	tmconfig "github.com/sei-protocol/sei-chain/sei-tendermint/config"
+)
+
+// readP2PConfig decodes a config.toml into a default Config: absent keys keep
+// the value already in the struct.
+func readP2PConfig(t *testing.T, body string) *tmconfig.P2PConfig {
+	t.Helper()
+
+	path := filepath.Join(t.TempDir(), "config.toml")
+	require.NoError(t, os.WriteFile(path, []byte(body), 0600))
+
+	v := viper.New()
+	v.SetConfigFile(path)
+	require.NoError(t, v.ReadInConfig())
+
+	cfg := tmconfig.DefaultConfig()
+	require.NoError(t, v.Unmarshal(cfg))
+	return cfg.P2P
+}
+
+// dial-interval is absent from the generated template (see checkConfig in
+// toml_test.go), so nothing else shows it is readable at all.
+func TestP2PPacingKnobsParseFromExistingConfig(t *testing.T) {
+	p2p := readP2PConfig(t, `
+[p2p]
+laddr = "tcp://0.0.0.0:26656"
+dial-interval = "5s"
+accept-interval = "20ms"
+`)
+
+	require.Equal(t, 5*time.Second, p2p.DialInterval)
+	require.Equal(t, 20*time.Millisecond, p2p.AcceptInterval)
+	require.NoError(t, p2p.ValidateBasic())
+}
+
+// TestP2PConfigPredatingPacingKnobsKeepsDefaults asserts a config.toml written
+// before these keys existed still parses to the defaults rather than to zero,
+// which rate.Every would read as "no pacing".
+func TestP2PConfigPredatingPacingKnobsKeepsDefaults(t *testing.T) {
+	p2p := readP2PConfig(t, `
+[p2p]
+laddr = "tcp://0.0.0.0:26656"
+`)
+
+	defaults := tmconfig.DefaultP2PConfig()
+	require.Equal(t, defaults.AcceptInterval, p2p.AcceptInterval)
+	require.Equal(t, defaults.DialInterval, p2p.DialInterval)
+	require.NotZero(t, p2p.AcceptInterval, "a zero accept-interval disables accept pacing entirely")
+	require.NoError(t, p2p.ValidateBasic())
+}

sei-tendermint/config/toml.go

@@ -309,6 +309,12 @@ allow-duplicate-ip = {{ .P2P.AllowDuplicateIP }}
 handshake-timeout = "{{ .P2P.HandshakeTimeout }}"
 dial-timeout = "{{ .P2P.DialTimeout }}"
 
+# How often the node accepts a new inbound connection. A larger interval paces
+# the accept loop more slowly; if the kernel accept backlog outpaces it, arriving
+# peers wait past handshake-timeout and the node silently stops acquiring inbound
+# peers. A value of 0 disables the limiter.
+accept-interval = "{{ .P2P.AcceptInterval }}"
+
 # Time to wait before flushing messages out on the connection
 # TODO: Remove once MConnConnection is removed.
 flush-throttle-timeout = "{{ .P2P.FlushThrottleTimeout }}"

sei-tendermint/config/toml_test.go

@@ -97,6 +97,25 @@ func checkConfig(t *testing.T, configFile string) {
 			t.Errorf("config file was not expected to contain %s", e)
 		}
 	}
+
+	// accept-interval is rendered deliberately: an accept rate too low to drain
+	// the kernel backlog silently stops a node acquiring inbound peers, and the
+	// template is where an operator looks. Keep it discoverable.
+	if !configContainsKey(configFile, "accept-interval") {
+		t.Errorf("config file was expected to contain accept-interval but did not")
+	}
+
+	// dial-interval remains an expert-only knob, left out of the generated
+	// template while still being parsed from existing config files.
+	// See TestP2PPacingKnobsParseFromExistingConfig.
+	var hiddenP2PElems = []string{
+		"dial-interval",
+	}
+	for _, e := range hiddenP2PElems {
+		if configContainsKey(configFile, e) {
+			t.Errorf("config file was not expected to contain %s", e)
+		}
+	}
 }
 
 func configContainsKey(configFile string, key string) bool {

sei-tendermint/internal/p2p/routeroptions.go

@@ -64,7 +64,8 @@ type RouterOptions struct {
 	// MaxDialRate limits the rate at which router is dialing peers. Defaults to 0.1/s.
 	MaxDialRate utils.Option[rate.Limit]
 
-	// MaxAcceptRate limits the rate at which router is accepting TCP connections. Defaults to 1/s.
+	// MaxAcceptRate limits the sustained rate at which router is accepting TCP
+	// connections; the limiter's burst is MaxConcurrentAccepts. Defaults to 100/s.
 	MaxAcceptRate utils.Option[rate.Limit]
 
 	// ResolveTimeout is the timeout for resolving NodeAddress URLs.
@@ -161,7 +162,11 @@ func (o *RouterOptions) maxDialRate() rate.Limit {
 }
 
 func (o *RouterOptions) maxAcceptRate() rate.Limit {
-	return o.MaxAcceptRate.Or(rate.Every(time.Second))
+	// The fallback is load-bearing on this branch: unlike main, RouterOptions.Validate
+	// here has no "rate must be > 0" guard, so a construction site that leaves the
+	// field unset silently inherits this value rather than failing. 1/s was too low
+	// to drain the listen backlog, which is the defect #3899 fixes.
+	return o.MaxAcceptRate.Or(rate.Every(10 * time.Millisecond))
 }
 
 func (o *RouterOptions) incomingConnectionWindow() time.Duration {

sei-tendermint/internal/p2p/routeroptions_test.go

@@ -0,0 +1,24 @@
+package p2p
+
+import (
+	"testing"
+	"time"
+
+	"golang.org/x/time/rate"
+
+	"github.com/sei-protocol/sei-chain/sei-tendermint/libs/utils"
+	"github.com/sei-protocol/sei-chain/sei-tendermint/libs/utils/require"
+)
+
+// Every test harness pins MaxAcceptRate/MaxDialRate to rate.Inf and node setup
+// always sets them, so these fallbacks are otherwise unexercised.
+func TestRouterOptionsPacingDefaults(t *testing.T) {
+	var o RouterOptions
+
+	require.Equal(t, rate.Every(10*time.Millisecond), o.maxAcceptRate())
+	require.Equal(t, rate.Every(10*time.Second), o.maxDialRate())
+
+	// An explicit value wins over the fallback.
+	o.MaxAcceptRate = utils.Some(rate.Inf)
+	require.Equal(t, rate.Inf, o.maxAcceptRate())
+}

sei-tendermint/node/setup.go

@@ -263,26 +263,22 @@ func buildGigaConfig(
 	}, nil
 }
 
-func createRouter(
-	p2pMetrics *p2p.Metrics,
-	nodeInfoProducer func() *types.NodeInfo,
-	nodeKey types.NodeKey,
-	validatorKey utils.Option[atypes.SecretKey],
-	cfg *config.Config,
-	app utils.Option[*proxy.Proxy],
-	genDoc *types.GenesisDoc,
-	dbProvider config.DBProvider,
-) (*p2p.Router, closer, error) {
-	closer := func() error { return nil }
-	ep, err := p2p.ResolveEndpoint(nodeKey.ID().AddressString(cfg.P2P.ListenAddress))
-	if err != nil {
-		return nil, closer, err
-	}
-	var privatePeerIDs []types.NodeID
-	for _, id := range tmstrings.SplitAndTrimEmpty(cfg.P2P.PrivatePeerIDs, ",", " ") {
-		privatePeerIDs = append(privatePeerIDs, types.NodeID(id))
-	}
+// pacingRate returns the rate limit for a configured pacing interval.
+func pacingRate(key string, interval time.Duration) (rate.Limit, error) {
+	// rate.Every maps every non-positive interval to rate.Inf. A configured 0 means
+	// "disable the limiter" and is honoured; a negative value is a typo that would
+	// silently disable pacing instead. ValidateBasic rejects it wherever it runs,
+	// but an already-deployed config never reaches ValidateBasic, so refuse it here
+	// too rather than letting the two paths disagree about the same input.
+	if interval < 0 {
+		return 0, fmt.Errorf("p2p %v must not be negative, got %v", key, interval)
+	}
+	return rate.Every(interval), nil
+}
 
+// p2pRouterOptions returns the router's connection budget and pacing, derived
+// from the p2p config.
+func p2pRouterOptions(cfg *config.Config, ep p2p.Endpoint, privatePeerIDs []types.NodeID) (*p2p.RouterOptions, error) {
 	// MaxConnections defaults to 64
 	maxConns := 64
 	if cfg.P2P.MaxConnections > 0 {
@@ -304,10 +300,19 @@ func createRouter(
 	connection.SendRate = cfg.P2P.SendRate
 	connection.RecvRate = cfg.P2P.RecvRate
 	connection.MaxPacketMsgPayloadSize = cfg.P2P.MaxPacketMsgPayloadSize
-	options := &p2p.RouterOptions{
+	dialRate, err := pacingRate("dial-interval", cfg.P2P.DialInterval)
+	if err != nil {
+		return nil, err
+	}
+	acceptRate, err := pacingRate("accept-interval", cfg.P2P.AcceptInterval)
+	if err != nil {
+		return nil, err
+	}
+	return &p2p.RouterOptions{
 		Endpoint:                      ep,
 		MaxIncomingConnectionAttempts: utils.Some(cfg.P2P.MaxIncomingConnectionAttempts),
-		MaxDialRate:                   utils.Some(rate.Every(cfg.P2P.DialInterval)),
+		MaxDialRate:                   utils.Some(dialRate),
+		MaxAcceptRate:                 utils.Some(acceptRate),
 		HandshakeTimeout:              utils.Some(cfg.P2P.HandshakeTimeout),
 		DialTimeout:                   utils.Some(cfg.P2P.DialTimeout),
 		PexOnHandshake:                cfg.P2P.PexReactor,
@@ -316,6 +321,32 @@ func createRouter(
 		MaxOutbound:                   utils.Some(maxOutbound),
 		MaxConcurrentAccepts:          utils.Some(maxInbound),
 		Connection:                    connection,
+	}, nil
+}
+
+func createRouter(
+	p2pMetrics *p2p.Metrics,
+	nodeInfoProducer func() *types.NodeInfo,
+	nodeKey types.NodeKey,
+	validatorKey utils.Option[atypes.SecretKey],
+	cfg *config.Config,
+	app utils.Option[*proxy.Proxy],
+	genDoc *types.GenesisDoc,
+	dbProvider config.DBProvider,
+) (*p2p.Router, closer, error) {
+	closer := func() error { return nil }
+	ep, err := p2p.ResolveEndpoint(nodeKey.ID().AddressString(cfg.P2P.ListenAddress))
+	if err != nil {
+		return nil, closer, err
+	}
+	var privatePeerIDs []types.NodeID
+	for _, id := range tmstrings.SplitAndTrimEmpty(cfg.P2P.PrivatePeerIDs, ",", " ") {
+		privatePeerIDs = append(privatePeerIDs, types.NodeID(id))
+	}
+
+	options, err := p2pRouterOptions(cfg, ep, privatePeerIDs)
+	if err != nil {
+		return nil, closer, err
 	}
 	if addr := cfg.P2P.ExternalAddress; addr != "" {
 		nodeAddr, err := p2p.ParseNodeAddress(nodeKey.ID().AddressString(addr))