Add flag to divert V2 traffic to new backend (#1734)

This is to help understand latency (especially website) on dev(/tbd new
test instance)

Currently only set to divert Node and Observation, since those are the
only ones with feature parity at the moment

Set to always use Spanner for local

Author: n-h-diaz

deploy/featureflags/local.yaml

@@ -3,3 +3,4 @@ flags:
   V3MirrorFraction: 1.0
   UseSpannerGraph: true
   UseStaleReads: true
+  V2DivertFraction: 1.0

internal/featureflags/featureflags.go

@@ -38,6 +38,8 @@ type Flags struct {
 	UseStaleReads bool `yaml:"UseStaleReads"`
 	// Whether to enable the embeddings resolver.
 	EnableEmbeddingsResolver bool `yaml:"EnableEmbeddingsResolver"`
+	// Fraction of V2 API requests to divert to the new dispatcher backend. Value from 0 to 1.0.
+	V2DivertFraction float64 `yaml:"V2DivertFraction"`
 }
 
 // setDefaultValues creates a new Flags struct with default values.
@@ -49,6 +51,7 @@ func setDefaultValues() *Flags {
 		SpannerGraphDatabase:     "",
 		UseStaleReads:            false,
 		EnableEmbeddingsResolver: false,
+		V2DivertFraction:         0.0,
 	}
 }
 
@@ -66,6 +69,12 @@ func (f *Flags) validateFlagValues() error {
 	if f.UseStaleReads && !f.UseSpannerGraph {
 		return fmt.Errorf("UseStaleReads requires UseSpannerGraph to be true")
 	}
+	if f.V2DivertFraction < 0 || f.V2DivertFraction > 1.0 {
+		return fmt.Errorf("V2DivertFraction must be between 0 and 1.0, got %f", f.V2DivertFraction)
+	}
+	if f.V2DivertFraction > 0 && !f.EnableV3 {
+		return fmt.Errorf("V2DivertFraction > 0 requires EnableV3 to be true")
+	}
 	return nil
 }
 

internal/server/handler_v2.go

@@ -18,12 +18,14 @@ package server
 import (
 	"context"
 	"log/slog"
+	"math/rand"
 	"time"
 
 	"github.com/datacommonsorg/mixer/internal/log"
 	"github.com/datacommonsorg/mixer/internal/merger"
 	pb "github.com/datacommonsorg/mixer/internal/proto"
 	pbv2 "github.com/datacommonsorg/mixer/internal/proto/v2"
+	"github.com/datacommonsorg/mixer/internal/server/datasources"
 	"github.com/datacommonsorg/mixer/internal/server/pagination"
 	"github.com/datacommonsorg/mixer/internal/server/statvar/search"
 	"github.com/datacommonsorg/mixer/internal/server/translator"
@@ -105,6 +107,11 @@ func (s *Server) V2Resolve(
 func (s *Server) V2Node(ctx context.Context, in *pbv2.NodeRequest) (
 	*pbv2.NodeResponse, error,
 ) {
+	if rand.Float64() < s.flags.V2DivertFraction {
+		slog.Info("V2Node request diverted to dispatcher backend", "request", in)
+		return s.dispatcher.Node(ctx, in, datasources.DefaultPageSize)
+	}
+
 	v2StartTime := time.Now()
 	errGroup, errCtx := errgroup.WithContext(ctx)
 	localRespChan := make(chan *pbv2.NodeResponse, 1)
@@ -260,6 +267,11 @@ func (s *Server) V2Event(
 func (s *Server) V2Observation(
 	ctx context.Context, in *pbv2.ObservationRequest,
 ) (*pbv2.ObservationResponse, error) {
+	if rand.Float64() < s.flags.V2DivertFraction {
+		slog.Info("V2Observation request diverted to dispatcher backend", "request", in)
+		return s.dispatcher.Observation(ctx, in)
+	}
+
 	v2StartTime := time.Now()
 
 	surface, toRemote := util.GetMetadata(ctx)