feat: client side load balancing for user transformations

go.mod

@@ -35,6 +35,7 @@ require (
 	github.com/apache/pulsar-client-go v0.14.0
 	github.com/araddon/dateparse v0.0.0-20210429162001-6b43995a97de
 	github.com/aws/aws-sdk-go v1.55.5
+	github.com/bufbuild/httplb v0.3.0
 	github.com/cenkalti/backoff v2.2.1+incompatible
 	github.com/cenkalti/backoff/v4 v4.3.0
 	github.com/confluentinc/confluent-kafka-go/v2 v2.6.1

go.sum

@@ -357,6 +357,8 @@ github.com/bsm/ginkgo/v2 v2.12.0/go.mod h1:SwYbGRRDovPVboqFv0tPTcG1sN61LM1Z4ARdb
 github.com/bsm/gomega v1.26.0/go.mod h1:JyEr/xRbxbtgWNi8tIEVPUYZ5Dzef52k01W3YH0H+O0=
 github.com/bsm/gomega v1.27.10 h1:yeMWxP2pV2fG3FgAODIY8EiRE3dy0aeFYt4l7wh6yKA=
 github.com/bsm/gomega v1.27.10/go.mod h1:JyEr/xRbxbtgWNi8tIEVPUYZ5Dzef52k01W3YH0H+O0=
+github.com/bufbuild/httplb v0.3.0 h1:sCMPD+89ydD3atcVareDsiv/kUT+pLHolENMoCGZJV8=
+github.com/bufbuild/httplb v0.3.0/go.mod h1:qDNs7dSFxIhKi/DA/rCCPVzbQfHs1JVxPMl9EvrbL4Q=
 github.com/buger/goterm v1.0.4 h1:Z9YvGmOih81P0FbVtEYTFF6YsSgxSUKEhf/f9bTMXbY=
 github.com/buger/goterm v1.0.4/go.mod h1:HiFWV3xnkolgrBV3mY8m0X0Pumt4zg4QhbdOzQtB8tE=
 github.com/bytedance/sonic v1.11.6/go.mod h1:LysEHSvpvDySVdC2f87zGWf6CIKJcAvqab1ZaiQtds4=

processor/transformer/transformer.go

@@ -7,6 +7,7 @@ import (
 	"context"
 	"fmt"
 	"io"
+	"net"
 	"net/http"
 	"os"
 	"runtime/trace"
@@ -15,6 +16,8 @@ import (
 	"sync"
 	"time"
 
+	"github.com/bufbuild/httplb"
+	"github.com/bufbuild/httplb/resolver"
 	"github.com/cenkalti/backoff"
 	jsoniter "github.com/json-iterator/go"
 	"github.com/samber/lo"
@@ -26,6 +29,7 @@ import (
 	backendconfig "github.com/rudderlabs/rudder-server/backend-config"
 	"github.com/rudderlabs/rudder-server/processor/integrations"
 	"github.com/rudderlabs/rudder-server/utils/httputil"
+	"github.com/rudderlabs/rudder-server/utils/sysUtils"
 	"github.com/rudderlabs/rudder-server/utils/types"
 	warehouseutils "github.com/rudderlabs/rudder-server/warehouse/utils"
 )
@@ -136,9 +140,9 @@ type Response struct {
 
 type Opt func(*handle)
 
-func WithClient(client *http.Client) Opt {
+func WithClient(client HTTPDoer) Opt {
 	return func(s *handle) {
-		s.client = client
+		s.httpClient = client
 	}
 }
 
@@ -149,6 +153,10 @@ type Transformer interface {
 	Validate(ctx context.Context, clientEvents []TransformerEvent, batchSize int) Response
 }
 
+type HTTPDoer interface {
+	Do(req *http.Request) (*http.Response, error)
+}
+
 // handle is the handle for this class
 type handle struct {
 	sentStat     stats.Measurement
@@ -159,7 +167,7 @@ type handle struct {
 	logger logger.Logger
 	stat   stats.Stats
 
-	client *http.Client
+	httpClient HTTPDoer
 
 	guardConcurrency chan struct{}
 
@@ -211,16 +219,41 @@ func NewTransformer(conf *config.Config, log logger.Logger, stat stats.Stats, op
 
 	trans.guardConcurrency = make(chan struct{}, trans.config.maxConcurrency)
 
-	if trans.client == nil {
-		trans.client = &http.Client{
-			Transport: &http.Transport{
-				DisableKeepAlives:   trans.config.disableKeepAlives,
-				MaxConnsPerHost:     trans.config.maxHTTPConnections,
-				MaxIdleConnsPerHost: trans.config.maxHTTPIdleConnections,
-				IdleConnTimeout:     trans.config.maxIdleConnDuration,
-			},
-			Timeout: trans.config.timeoutDuration,
-		}
+	clientType := conf.GetString("Transformer.Client.type", "stdlib")
+
+	transport := &http.Transport{
+		DisableKeepAlives:   trans.config.disableKeepAlives,
+		MaxConnsPerHost:     trans.config.maxHTTPConnections,
+		MaxIdleConnsPerHost: trans.config.maxHTTPIdleConnections,
+		IdleConnTimeout:     trans.config.maxIdleConnDuration,
+	}
+	client := &http.Client{
+		Transport: transport,
+		Timeout:   trans.config.timeoutDuration,
+	}
+
+	switch clientType {
+	case "stdlib":
+		trans.httpClient = client
+	case "recycled":
+		trans.httpClient = sysUtils.NewRecycledHTTPClient(func() *http.Client {
+			return client
+		}, config.GetDuration("Transformer.Client.ttl", 120, time.Second))
+	case "httplb":
+		trans.httpClient = httplb.NewClient(
+			httplb.WithTransport("http", &HTTPLBTransport{
+				Transport: transport,
+			}),
+			httplb.WithResolver(
+				resolver.NewDNSResolver(
+					net.DefaultResolver,
+					resolver.PreferIPv6,
+					config.GetDuration("Transformer.Client.ttl", 120, time.Second), // TTL value
+				),
+			),
+		)
+	default:
+		panic(fmt.Sprintf("unknown transformer client type: %s", clientType))
 	}
 
 	for _, opt := range opts {
@@ -245,6 +278,14 @@ func (trans *handle) Validate(ctx context.Context, clientEvents []TransformerEve
 	return trans.transform(ctx, clientEvents, trans.trackingPlanValidationURL(), batchSize, trackingPlanValidationStage)
 }
 
+type HTTPLBTransport struct {
+	*http.Transport
+}
+
+func (t *HTTPLBTransport) NewRoundTripper(scheme, target string, config httplb.TransportConfig) httplb.RoundTripperResult {
+	return httplb.RoundTripperResult{RoundTripper: t.Transport, Close: t.CloseIdleConnections}
+}
+
 func (trans *handle) transform(
 	ctx context.Context,
 	clientEvents []TransformerEvent,
@@ -474,7 +515,7 @@ func (trans *handle) doPost(ctx context.Context, rawJSON []byte, url, stage stri
 				// Header to let transformer know that the client understands event filter code
 				req.Header.Set("X-Feature-Filter-Code", "?1")
 
-				resp, reqErr = trans.client.Do(req)
+				resp, reqErr = trans.httpClient.Do(req)
 			})
 			trans.stat.NewTaggedStat("processor.transformer_request_time", stats.TimerType, tags).SendTiming(time.Since(requestStartTime))
 			if reqErr != nil {
@@ -495,7 +536,6 @@ func (trans *handle) doPost(ctx context.Context, rawJSON []byte, url, stage stri
 			retryCount++
 			trans.logger.Warnn(
 				"JS HTTP connection error",
-				logger.NewStringField("URL", url),
 				logger.NewErrorField(err),
 				logger.NewIntField("attempts", int64(retryCount)),
 			)