perf: expose and reduce message publish latency

This commit is contained in:
richjohnrambo
2026-09-27 21:08:35 +08:00
parent 8e778726e0
commit bde02e8eac
15 changed files with 274 additions and 98 deletions
+41 -30
View File
@@ -10,35 +10,38 @@ import (
)
type CombinedReport struct {
RunID string `json:"run_id"`
Scenario string `json:"scenario"`
GeneratedAt time.Time `json:"generated_at"`
Agents []AgentSummary `json:"agents"`
ExpectedConnections int `json:"expected_connections"`
ReadyConnections int `json:"ready_connections"`
PublishAttempted int64 `json:"publish_attempted"`
PublishAccepted int64 `json:"publish_accepted"`
PublishRejected int64 `json:"publish_rejected"`
PublishErrors map[string]int64 `json:"publish_errors,omitempty"`
PublishSuccessRate float64 `json:"publish_success_rate"`
AckP95MS float64 `json:"ack_p95_ms"`
DeliveryP95MS float64 `json:"delivery_p95_ms"`
ExpectedDeliveries int64 `json:"expected_deliveries"`
ReceivedDeliveries int64 `json:"received_deliveries"`
RealtimeLost int64 `json:"realtime_lost"`
RealtimeLostStored int64 `json:"realtime_lost_but_persisted"`
ArrivalRate float64 `json:"arrival_rate"`
LossRate float64 `json:"loss_rate"`
Duplicates int64 `json:"duplicates"`
OutOfOrder int64 `json:"out_of_order"`
OrderingViolations []OrderingViolation `json:"ordering_violations,omitempty"`
Disconnects int64 `json:"disconnects"`
HistoryChecked int64 `json:"history_checked"`
HistoryFound int64 `json:"history_found"`
PersistenceLost int64 `json:"persistence_lost"`
Missing []MissingDelivery `json:"missing,omitempty"`
Passed bool `json:"passed"`
FailureReasons []string `json:"failure_reasons,omitempty"`
RunID string `json:"run_id"`
Scenario string `json:"scenario"`
GeneratedAt time.Time `json:"generated_at"`
Agents []AgentSummary `json:"agents"`
ExpectedConnections int `json:"expected_connections"`
ReadyConnections int `json:"ready_connections"`
PublishAttempted int64 `json:"publish_attempted"`
PublishAccepted int64 `json:"publish_accepted"`
PublishRejected int64 `json:"publish_rejected"`
PublishErrors map[string]int64 `json:"publish_errors,omitempty"`
PublishSuccessRate float64 `json:"publish_success_rate"`
AckP95MS float64 `json:"ack_p95_ms"`
DeliveryP95MS float64 `json:"delivery_p95_ms"`
DeliveryLocalP95MS float64 `json:"delivery_local_p95_ms"`
DeliveryCrossP95MS float64 `json:"delivery_cross_machine_p95_ms"`
CrossClockSkewSamples int64 `json:"cross_machine_negative_latency_samples"`
ExpectedDeliveries int64 `json:"expected_deliveries"`
ReceivedDeliveries int64 `json:"received_deliveries"`
RealtimeLost int64 `json:"realtime_lost"`
RealtimeLostStored int64 `json:"realtime_lost_but_persisted"`
ArrivalRate float64 `json:"arrival_rate"`
LossRate float64 `json:"loss_rate"`
Duplicates int64 `json:"duplicates"`
OutOfOrder int64 `json:"out_of_order"`
OrderingViolations []OrderingViolation `json:"ordering_violations,omitempty"`
Disconnects int64 `json:"disconnects"`
HistoryChecked int64 `json:"history_checked"`
HistoryFound int64 `json:"history_found"`
PersistenceLost int64 `json:"persistence_lost"`
Missing []MissingDelivery `json:"missing,omitempty"`
Passed bool `json:"passed"`
FailureReasons []string `json:"failure_reasons,omitempty"`
}
func runMerge(aPath, bPath, output string) error {
@@ -79,6 +82,7 @@ func runMerge(aPath, bPath, output string) error {
report.HistoryChecked += summary.HistoryChecked
report.HistoryFound += summary.HistoryFound
report.PersistenceLost += summary.PersistenceLost
report.CrossClockSkewSamples += summary.CrossClockSkewSamples
report.Missing = append(report.Missing, summary.Missing...)
if len(summary.Errors) > 0 {
report.FailureReasons = append(report.FailureReasons, fmt.Sprintf("agent %s errors=%d", summary.Machine, len(summary.Errors)))
@@ -88,7 +92,14 @@ func runMerge(aPath, bPath, output string) error {
report.PublishSuccessRate = float64(report.PublishAccepted) / float64(report.PublishAttempted)
}
report.AckP95MS = mergedAckP95(report.Agents)
report.DeliveryP95MS = mergedHistogramP95(report.Agents, func(summary AgentSummary) HistogramSnapshot { return summary.DeliveryLatency })
report.DeliveryLocalP95MS = mergedHistogramP95(report.Agents, func(summary AgentSummary) HistogramSnapshot { return summary.DeliveryLatencyLocal })
report.DeliveryCrossP95MS = mergedHistogramP95(report.Agents, func(summary AgentSummary) HistogramSnapshot { return summary.DeliveryLatencyCross })
// delivery_p95_ms remains the compatibility field, but now uses only same-clock
// samples. Cross-machine wall clocks cannot produce a valid one-way latency.
report.DeliveryP95MS = report.DeliveryLocalP95MS
if report.DeliveryP95MS == 0 {
report.DeliveryP95MS = mergedHistogramP95(report.Agents, func(summary AgentSummary) HistogramSnapshot { return summary.DeliveryLatency })
}
if report.ExpectedDeliveries > 0 {
report.ArrivalRate = float64(report.ReceivedDeliveries) / float64(report.ExpectedDeliveries)
report.LossRate = float64(report.RealtimeLost) / float64(report.ExpectedDeliveries)