chore(*): use proper retry

btcclient/query.go

@@ -3,7 +3,7 @@ package btcclient
 import (
 	"fmt"
 
-	"github.com/babylonlabs-io/babylon/types/retry"
+	"github.com/avast/retry-go/v4"
 	"github.com/btcsuite/btcd/btcjson"
 	"github.com/btcsuite/btcd/chaincfg/chainhash"
 	"github.com/btcsuite/btcd/wire"
@@ -66,13 +66,16 @@ func (c *Client) getBestBlockHashWithRetry() (*chainhash.Hash, error) {
 		err       error
 	)
 
-	if err := retry.Do(c.retrySleepTime, c.maxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		blockHash, err = c.GetBestBlockHash()
 		if err != nil {
 			return err
 		}
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(c.retrySleepTime),
+		retry.MaxDelay(c.maxRetrySleepTime),
+	); err != nil {
 		c.logger.Debug(
 			"failed to query the best block hash", zap.Error(err))
 	}
@@ -86,13 +89,16 @@ func (c *Client) getBlockHashWithRetry(height uint64) (*chainhash.Hash, error) {
 		err       error
 	)
 
-	if err := retry.Do(c.retrySleepTime, c.maxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		blockHash, err = c.GetBlockHash(int64(height))
 		if err != nil {
 			return err
 		}
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(c.retrySleepTime),
+		retry.MaxDelay(c.maxRetrySleepTime),
+	); err != nil {
 		c.logger.Debug(
 			"failed to query the block hash", zap.Uint64("height", height), zap.Error(err))
 	}
@@ -106,13 +112,16 @@ func (c *Client) getBlockWithRetry(hash *chainhash.Hash) (*wire.MsgBlock, error)
 		err   error
 	)
 
-	if err := retry.Do(c.retrySleepTime, c.maxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		block, err = c.GetBlock(hash)
 		if err != nil {
 			return err
 		}
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(c.retrySleepTime),
+		retry.MaxDelay(c.maxRetrySleepTime),
+	); err != nil {
 		c.logger.Debug(
 			"failed to query the block", zap.String("hash", hash.String()), zap.Error(err))
 	}
@@ -126,13 +135,16 @@ func (c *Client) getBlockVerboseWithRetry(hash *chainhash.Hash) (*btcjson.GetBlo
 		err          error
 	)
 
-	if err := retry.Do(c.retrySleepTime, c.maxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		blockVerbose, err = c.GetBlockVerbose(hash)
 		if err != nil {
 			return err
 		}
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(c.retrySleepTime),
+		retry.MaxDelay(c.maxRetrySleepTime),
+	); err != nil {
 		c.logger.Debug(
 			"failed to query the block verbose", zap.String("hash", hash.String()), zap.Error(err))
 	}

config/common.go

@@ -10,6 +10,7 @@ import (
 const (
 	defaultRetrySleepTime    = 5 * time.Second
 	defaultMaxRetrySleepTime = 5 * time.Minute
+	defaultMaxRetryTimes     = 5
 )
 
 // CommonConfig defines the server's basic configuration
@@ -23,6 +24,8 @@ type CommonConfig struct {
 	// Maximum backoff interval between retries. Exponential backoff leads to interval increase.
 	// This value is the cap of the interval, when exceeded the retries stop.
 	MaxRetrySleepTime time.Duration `mapstructure:"max-retry-sleep-time"`
+	// The max number of retries in case of a failure
+	MaxRetryTimes uint `mapstructure:"max-retry-times"`
 }
 
 func isOneOf(v string, list []string) bool {
@@ -60,5 +63,6 @@ func DefaultCommonConfig() CommonConfig {
 		LogLevel:          "debug",
 		RetrySleepTime:    defaultRetrySleepTime,
 		MaxRetrySleepTime: defaultMaxRetrySleepTime,
+		MaxRetryTimes:     defaultMaxRetryTimes,
 	}
 }

go.mod

@@ -6,7 +6,7 @@ require (
 	cosmossdk.io/errors v1.0.1
 	cosmossdk.io/log v1.3.1
 	cosmossdk.io/math v1.3.0
-	github.com/avast/retry-go/v4 v4.5.1
+	github.com/avast/retry-go/v4 v4.6.0
 	github.com/babylonlabs-io/babylon v0.9.1
 	github.com/boljen/go-bitmap v0.0.0-20151001105940-23cd2fb0ce7d
 	github.com/btcsuite/btcd v0.24.2

go.sum

@@ -275,8 +275,8 @@ github.com/armon/consul-api v0.0.0-20180202201655-eb2c6b5be1b6/go.mod h1:grANhF5
 github.com/armon/go-metrics v0.0.0-20180917152333-f0300d1749da/go.mod h1:Q73ZrmVTwzkszR9V5SSuryQ31EELlFMUz1kKyl939pY=
 github.com/armon/go-radix v0.0.0-20180808171621-7fddfc383310/go.mod h1:ufUuZ+zHj4x4TnLV4JWEpy2hxWSpsRywHrMgIH9cCH8=
 github.com/aryann/difflib v0.0.0-20170710044230-e206f873d14a/go.mod h1:DAHtR1m6lCRdSC2Tm3DSWRPvIPr6xNKyeHdqDQSQT+A=
-github.com/avast/retry-go/v4 v4.5.1 h1:AxIx0HGi4VZ3I02jr78j5lZ3M6x1E0Ivxa6b0pUUh7o=
-github.com/avast/retry-go/v4 v4.5.1/go.mod h1:/sipNsvNB3RRuT5iNcb6h73nw3IBmXJ/H3XrCQYSOpc=
+github.com/avast/retry-go/v4 v4.6.0 h1:K9xNA+KeB8HHc2aWFuLb25Offp+0iVRXEvFx8IinRJA=
+github.com/avast/retry-go/v4 v4.6.0/go.mod h1:gvWlPhBVsvBbLkVGDg/KwvBv0bEkCOLRRSHKIr2PyOE=
 github.com/aws/aws-lambda-go v1.13.3/go.mod h1:4UKl9IzQMoD+QF79YdCuzCwp8VbmG4VAQwij/eHl5CU=
 github.com/aws/aws-sdk-go v1.27.0/go.mod h1:KmX6BPdI08NWTb3/sm4ZGu5ShLoqVDhKgpiN924inxo=
 github.com/aws/aws-sdk-go v1.44.122/go.mod h1:y4AeaBuwd2Lk+GepC1E9v0qOiTws0MIWAX4oIKwKHZo=

monitor/liveness_checker_test.go

@@ -1,20 +1,18 @@
 package monitor_test
 
 import (
-	"math/rand"
-	"testing"
-
 	bbndatagen "github.com/babylonlabs-io/babylon/testutil/datagen"
 	bbntypes "github.com/babylonlabs-io/babylon/types"
 	btclctypes "github.com/babylonlabs-io/babylon/x/btclightclient/types"
 	monitortypes "github.com/babylonlabs-io/babylon/x/monitor/types"
-	"github.com/golang/mock/gomock"
-	"github.com/stretchr/testify/require"
-
 	"github.com/babylonlabs-io/vigilante/config"
 	"github.com/babylonlabs-io/vigilante/monitor"
 	"github.com/babylonlabs-io/vigilante/testutil/datagen"
 	"github.com/babylonlabs-io/vigilante/types"
+	"github.com/golang/mock/gomock"
+	"github.com/stretchr/testify/require"
+	"math/rand"
+	"testing"
 )
 
 func FuzzLivenessChecker(f *testing.F) {
@@ -33,6 +31,7 @@ func FuzzLivenessChecker(f *testing.F) {
 			ComCfg: &config.CommonConfig{
 				RetrySleepTime:    1,
 				MaxRetrySleepTime: 0,
+				MaxRetryTimes:     1,
 			},
 			BBNQuerier: mockBabylonClient,
 		}

monitor/query.go

@@ -3,7 +3,7 @@ package monitor
 import (
 	"fmt"
 
-	"github.com/babylonlabs-io/babylon/types/retry"
+	"github.com/avast/retry-go/v4"
 	btclctypes "github.com/babylonlabs-io/babylon/x/btclightclient/types"
 	ckpttypes "github.com/babylonlabs-io/babylon/x/checkpointing/types"
 	epochingtypes "github.com/babylonlabs-io/babylon/x/epoching/types"
@@ -61,15 +61,18 @@ func (m *Monitor) FindTipConfirmedEpoch() (uint64, error) {
 func (m *Monitor) queryCurrentEpochWithRetry() (*epochingtypes.QueryCurrentEpochResponse, error) {
 	var currentEpochRes epochingtypes.QueryCurrentEpochResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.CurrentEpoch()
 		if err != nil {
 			return err
 		}
 
 		currentEpochRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the current epoch", zap.Error(err))
 
@@ -82,15 +85,18 @@ func (m *Monitor) queryCurrentEpochWithRetry() (*epochingtypes.QueryCurrentEpoch
 func (m *Monitor) queryRawCheckpointWithRetry(epoch uint64) (*ckpttypes.QueryRawCheckpointResponse, error) {
 	var rawCheckpointRes ckpttypes.QueryRawCheckpointResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.RawCheckpoint(epoch)
 		if err != nil {
 			return err
 		}
 
 		rawCheckpointRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the raw checkpoint", zap.Error(err))
 
@@ -103,15 +109,18 @@ func (m *Monitor) queryRawCheckpointWithRetry(epoch uint64) (*ckpttypes.QueryRaw
 func (m *Monitor) queryBlsPublicKeyListWithRetry(epoch uint64) (*ckpttypes.QueryBlsPublicKeyListResponse, error) {
 	var blsPublicKeyListRes ckpttypes.QueryBlsPublicKeyListResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.BlsPublicKeyList(epoch, nil)
 		if err != nil {
 			return err
 		}
 
 		blsPublicKeyListRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the BLS public key list", zap.Error(err))
 
@@ -124,15 +133,18 @@ func (m *Monitor) queryBlsPublicKeyListWithRetry(epoch uint64) (*ckpttypes.Query
 func (m *Monitor) queryEndedEpochBTCHeightWithRetry(epoch uint64) (*monitortypes.QueryEndedEpochBtcHeightResponse, error) {
 	var endedEpochBTCHeightRes monitortypes.QueryEndedEpochBtcHeightResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.EndedEpochBTCHeight(epoch)
 		if err != nil {
 			return err
 		}
 
 		endedEpochBTCHeightRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the ended epoch BTC height", zap.Error(err))
 
@@ -145,15 +157,19 @@ func (m *Monitor) queryEndedEpochBTCHeightWithRetry(epoch uint64) (*monitortypes
 func (m *Monitor) queryReportedCheckpointBTCHeightWithRetry(hashStr string) (*monitortypes.QueryReportedCheckpointBtcHeightResponse, error) {
 	var reportedCheckpointBtcHeightRes monitortypes.QueryReportedCheckpointBtcHeightResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.ReportedCheckpointBTCHeight(hashStr)
 		if err != nil {
 			return err
 		}
 
 		reportedCheckpointBtcHeightRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+		retry.Attempts(m.ComCfg.MaxRetryTimes),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the reported checkpoint BTC height", zap.Error(err))
 
@@ -166,15 +182,18 @@ func (m *Monitor) queryReportedCheckpointBTCHeightWithRetry(hashStr string) (*mo
 func (m *Monitor) queryBTCHeaderChainTipWithRetry() (*btclctypes.QueryTipResponse, error) {
 	var btcHeaderChainTipRes btclctypes.QueryTipResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.BTCHeaderChainTip()
 		if err != nil {
 			return err
 		}
 
 		btcHeaderChainTipRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the BTC header chain tip", zap.Error(err))
 
@@ -187,15 +206,18 @@ func (m *Monitor) queryBTCHeaderChainTipWithRetry() (*btclctypes.QueryTipRespons
 func (m *Monitor) queryContainsBTCBlockWithRetry(blockHash *chainhash.Hash) (*btclctypes.QueryContainsBytesResponse, error) {
 	var containsBTCBlockRes btclctypes.QueryContainsBytesResponse
 
-	if err := retry.Do(m.ComCfg.RetrySleepTime, m.ComCfg.MaxRetrySleepTime, func() error {
+	if err := retry.Do(func() error {
 		res, err := m.BBNQuerier.ContainsBTCBlock(blockHash)
 		if err != nil {
 			return err
 		}
 
 		containsBTCBlockRes = *res
 		return nil
-	}); err != nil {
+	},
+		retry.Delay(m.ComCfg.RetrySleepTime),
+		retry.MaxDelay(m.ComCfg.MaxRetrySleepTime),
+	); err != nil {
 		m.logger.Debug(
 			"failed to query the contains BTC block", zap.Error(err))
 

reporter/reporter.go

@@ -7,8 +7,8 @@ import (
 	"sync"
 	"time"
 
+	"github.com/avast/retry-go/v4"
 	"github.com/babylonlabs-io/babylon/btctxformatter"
-	"github.com/babylonlabs-io/babylon/types/retry"
 	btcctypes "github.com/babylonlabs-io/babylon/x/btccheckpoint/types"
 	"github.com/babylonlabs-io/vigilante/btcclient"
 	"github.com/babylonlabs-io/vigilante/config"
@@ -57,10 +57,13 @@ func New(
 		btccParamsRes *btcctypes.QueryParamsResponse
 		err           error
 	)
-	err = retry.Do(retrySleepTime, maxRetrySleepTime, func() error {
+	err = retry.Do(func() error {
 		btccParamsRes, err = babylonClient.BTCCheckpointParams()
 		return err
-	})
+	},
+		retry.Delay(retrySleepTime),
+		retry.MaxDelay(maxRetrySleepTime),
+	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to get BTC Checkpoint parameters: %w", err)
 	}

reporter/utils.go

@@ -7,7 +7,7 @@ import (
 
 	pv "github.com/cosmos/relayer/v2/relayer/provider"
 
-	"github.com/babylonlabs-io/babylon/types/retry"
+	"github.com/avast/retry-go/v4"
 	btcctypes "github.com/babylonlabs-io/babylon/x/btccheckpoint/types"
 	btclctypes "github.com/babylonlabs-io/babylon/x/btclightclient/types"
 	"github.com/babylonlabs-io/vigilante/types"
@@ -33,10 +33,13 @@ func (r *Reporter) getHeaderMsgsToSubmit(signer string, ibs []*types.IndexedBloc
 	for i, header := range ibs {
 		blockHash := header.BlockHash()
 		var res *btclctypes.QueryContainsBytesResponse
-		err = retry.Do(r.retrySleepTime, r.maxRetrySleepTime, func() error {
+		err = retry.Do(func() error {
 			res, err = r.babylonClient.ContainsBTCBlock(&blockHash)
 			return err
-		})
+		},
+			retry.Delay(r.retrySleepTime),
+			retry.MaxDelay(r.maxRetrySleepTime),
+		)
 		if err != nil {
 			return nil, err
 		}
@@ -57,7 +60,7 @@ func (r *Reporter) getHeaderMsgsToSubmit(signer string, ibs []*types.IndexedBloc
 
 	blockChunks := chunkBy(ibsToSubmit, int(r.Cfg.MaxHeadersInMsg))
 
-	headerMsgsToSubmit := []*btclctypes.MsgInsertHeaders{}
+	headerMsgsToSubmit := make([]*btclctypes.MsgInsertHeaders, 0, len(blockChunks))
 
 	for _, ibChunk := range blockChunks {
 		msgInsertHeaders := types.NewMsgInsertHeaders(signer, ibChunk)
@@ -69,14 +72,17 @@ func (r *Reporter) getHeaderMsgsToSubmit(signer string, ibs []*types.IndexedBloc
 
 func (r *Reporter) submitHeaderMsgs(msg *btclctypes.MsgInsertHeaders) error {
 	// submit the headers
-	err := retry.Do(r.retrySleepTime, r.maxRetrySleepTime, func() error {
+	err := retry.Do(func() error {
 		res, err := r.babylonClient.InsertHeaders(context.Background(), msg)
 		if err != nil {
 			return err
 		}
 		r.logger.Infof("Successfully submitted %d headers to Babylon with response code %v", len(msg.Headers), res.Code)
 		return nil
-	})
+	},
+		retry.Delay(r.retrySleepTime),
+		retry.MaxDelay(r.maxRetrySleepTime),
+	)
 	if err != nil {
 		r.metrics.FailedHeadersCounter.Add(float64(len(msg.Headers)))
 		return fmt.Errorf("failed to submit headers: %w", err)