Weights now uses trace_id to pick a choice (#44)

* choice with the same weights are now consistent within the same trace_id

* propagate errors to parent spans

CHANGELOG.md

@@ -4,9 +4,14 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [Unreleased](https://github.com/lightstep/telemetry-generator/compare/v0.11.11...HEAD)
+## [Unreleased](https://github.com/lightstep/telemetry-generator/compare/v0.11.12...HEAD)
 
 
+## [0.11.12](https://github.com/lightstep/telemetry-generator/compare/v0.11.11...v0.11.12) - 2023-3-03
+### Changed
+* Weights the last two digits of a trace_id to generate their randomness.
+* Errors are now propagated up to the parent span. 
+
 ## [0.11.11](https://github.com/lightstep/telemetry-generator/compare/v0.11.10...v0.11.11) - 2023-3-02
 ### Fixed
 * Fixed a bug where multiple traces were being created with the same trace_id and span_id.

README.md

@@ -119,6 +119,7 @@ version for the 10 most recent telemetry-generator versions.
 
 | Telemetry Generator | OpenTelemetry Collector |
 |---------------------|-------------------------|
+| v0.11.12 | v0.69.1 |
 | v0.11.11 | v0.69.1 |
 | v0.11.10 | v0.69.1 |
 | v0.11.9 | v0.68.0 |

VERSION

@@ -1 +1 @@
-0.11.11
+0.11.12

generatorreceiver/internal/generator/trace_generator.go

@@ -75,7 +75,7 @@ func (g *TraceGenerator) createSpanForServiceRouteCall(traces *ptrace.Traces, se
 
  resource.Attributes().PutStr(string(semconv.ServiceNameKey), serviceTier.ServiceName)
 
- resourceAttributeSet := serviceTier.GetResourceAttributeSet()
+ resourceAttributeSet := serviceTier.GetResourceAttributeSet(traceId)
  attrs := resource.Attributes()
  resourceAttributeSet.GetAttributes().InsertTags(&attrs)
 
@@ -92,7 +92,7 @@ func (g *TraceGenerator) createSpanForServiceRouteCall(traces *ptrace.Traces, se
  span.SetKind(ptrace.SpanKindServer)
  span.Attributes().PutStr("load_generator.seq_num", fmt.Sprintf("%v", g.sequenceNumber))
 
- ts := serviceTier.GetTagSet(routeName) // ts is single TagSet consisting of tags from the service AND route
+ ts := serviceTier.GetTagSet(routeName, traceId) // ts is single TagSet consisting of tags from the service AND route
  attr := span.Attributes()
  ts.Tags.InsertTags(&attr) // add service and route tags to span attributes
 
@@ -106,11 +106,16 @@ func (g *TraceGenerator) createSpanForServiceRouteCall(traces *ptrace.Traces, se
  // TODO: this is still a bit weird - we're calling each downstream route
  // after a sample of the current route's latency, which doesn't really
  // make sense - but maybe it's realistic enough?
- endTime := startTimeNanos + route.SampleLatency()
+ endTime := startTimeNanos + route.SampleLatency(traceId)
  for _, c := range route.DownstreamCalls {
- var childStartTimeNanos = startTimeNanos + route.SampleLatency()
+ var childStartTimeNanos = startTimeNanos + route.SampleLatency(traceId)
 
  childSpan := g.createSpanForServiceRouteCall(traces, c.Service, c.Route, childStartTimeNanos, traceId, newSpanId)
+ val, ok := childSpan.Attributes().Get("error")
+ if ok {
+ errorAttr := span.Attributes().PutEmpty("error")
+ val.CopyTo(errorAttr)
+ }
  endTime = Max(endTime, int64(childSpan.EndTimestamp()))
  }
 

generatorreceiver/internal/topology/latency_percentiles.go

@@ -1,9 +1,12 @@
 package topology
 
 import (
- "github.com/lightstep/telemetry-generator/generatorreceiver/internal/flags"
  "math/rand"
  "time"
+
+ "go.opentelemetry.io/collector/pdata/pcommon"
+
+ "github.com/lightstep/telemetry-generator/generatorreceiver/internal/flags"
 )
 
 type LatencyPercentiles struct {
@@ -21,6 +24,7 @@ type LatencyPercentiles struct {
  p999 time.Duration
  p100 time.Duration
  }
+ EmbeddedWeight `json:",inline" yaml:",inline"`
  flags.EmbeddedFlags `json:",inline" yaml:",inline"`
 }
 
@@ -80,7 +84,7 @@ func (l *LatencyPercentiles) loadDurations() error {
 
 type LatencyConfigs []*LatencyPercentiles
 
-func (lcfg *LatencyConfigs) Sample() int64 {
+func (lcfg *LatencyConfigs) Sample(traceID pcommon.TraceID) int64 {
  var defaultCfg *LatencyPercentiles
  var enabled []*LatencyPercentiles
  for _, cfg := range *lcfg {
@@ -91,7 +95,7 @@ func (lcfg *LatencyConfigs) Sample() int64 {
  }
  }
  if len(enabled) > 0 {
- return enabled[rand.Intn(len(enabled))].Sample()
+ return pickBasedOnWeight(enabled, traceID).Sample()
  }
  return defaultCfg.Sample()
 }

generatorreceiver/internal/topology/pickable.go

@@ -1,6 +1,10 @@
 package topology
 
-import "math/rand"
+import (
+ "encoding/binary"
+
+ "go.opentelemetry.io/collector/pdata/pcommon"
+)
 
 type Pickable interface { // currently TagSet and ResourceAttributeSet satisfy this interface
  GetWeight() float64
@@ -15,7 +19,7 @@ func (w EmbeddedWeight) GetWeight() float64 {
  return w.Weight
 }
 
-func pickBasedOnWeight[P Pickable](ps []P) P {
+func pickBasedOnWeight[P Pickable](ps []P, traceID pcommon.TraceID) P {
  var activeSets []P
  totalWeight := 0.0
  for _, set := range ps {
@@ -25,7 +29,14 @@ func pickBasedOnWeight[P Pickable](ps []P) P {
  }
  }
 
- choice := rand.Float64() * totalWeight
+ // Take out last 8 bytes from trace id
+ secondHalf := traceID[8:16]
+ // Transform them into a uint64
+ traceUint := binary.BigEndian.Uint64(secondHalf)
+ // Use the last two digits as percentage.
+ choice := float64(traceUint%100) / 100.0
+
+ choice = choice * totalWeight
  current := 0.0
  for _, set := range activeSets {
  current += set.GetWeight()

generatorreceiver/internal/topology/service_route.go

@@ -4,6 +4,8 @@ import (
  "fmt"
  "math/rand"
 
+ "go.opentelemetry.io/collector/pdata/pcommon"
+
  "github.com/lightstep/telemetry-generator/generatorreceiver/internal/flags"
 )
 
@@ -77,10 +79,10 @@ func (r *ServiceRoute) load(route string) error {
  return nil
 }
 
-func (r *ServiceRoute) SampleLatency() int64 {
+func (r *ServiceRoute) SampleLatency(traceID pcommon.TraceID) int64 {
  if r.LatencyConfigs == nil {
  return rand.Int63n(r.MaxLatencyMillis * 1000000)
  } else {
- return r.LatencyConfigs.Sample()
+ return r.LatencyConfigs.Sample(traceID)
  }
 }

generatorreceiver/internal/topology/service_tier.go

@@ -2,6 +2,8 @@ package topology
 
 import (
  "fmt"
+
+ "go.opentelemetry.io/collector/pdata/pcommon"
 )
 
 type ServiceTier struct {
@@ -12,9 +14,9 @@ type ServiceTier struct {
  Metrics []Metric `json:"metrics" yaml:"metrics"`
 }
 
-func (st *ServiceTier) GetTagSet(routeName string) TagSet {
- serviceTagSet := pickBasedOnWeight(st.TagSets)
- routeTagSet := pickBasedOnWeight(st.GetRoute(routeName).TagSets)
+func (st *ServiceTier) GetTagSet(routeName string, traceID pcommon.TraceID) TagSet {
+ serviceTagSet := pickBasedOnWeight(st.TagSets, traceID)
+ routeTagSet := pickBasedOnWeight(st.GetRoute(routeName).TagSets, traceID)
 
  combinedTags := TagMap{}
  for k, v := range serviceTagSet.Tags {
@@ -30,9 +32,9 @@ func (st *ServiceTier) GetTagSet(routeName string) TagSet {
  }
 }
 
-func (st *ServiceTier) GetResourceAttributeSet() ResourceAttributeSet {
+func (st *ServiceTier) GetResourceAttributeSet(traceID pcommon.TraceID) ResourceAttributeSet {
  // TODO: also support resource attributes on routes
- return pickBasedOnWeight(st.ResourceAttributeSets)
+ return pickBasedOnWeight(st.ResourceAttributeSets, traceID)
 }
 
 func (st *ServiceTier) GetRoute(routeName string) *ServiceRoute {