feat: record task attempt chains

2026-05-11 22:39:45 +08:00
parent 4c2de4b4c9
commit 0049b246c1
9 changed files with 492 additions and 9 deletions
@@ -40,7 +40,8 @@ func buildSuccessRecord(task store.GatewayTask, user *auth.User, body map[string
 }

 func taskMetrics(task store.GatewayTask, user *auth.User, body map[string]any, candidate store.RuntimeModelCandidate, response clients.Response, simulated bool) map[string]any {
-	metrics := map[string]any{
+	metrics := attemptMetrics(candidate, 0, simulated)
+	for key, value := range map[string]any{
 		"kind":            task.Kind,
 		"runMode":         task.RunMode,
 		"requestedModel":  task.Model,
@@ -57,6 +58,8 @@ func taskMetrics(task store.GatewayTask, user *auth.User, body map[string]any, c
 		"queueKey":        candidate.QueueKey,
 		"requestId":       response.RequestID,
 		"simulated":       simulated,
+	} {
+		metrics[key] = value
 	}
 	if user != nil {
 		metrics["apiKeyId"] = user.APIKeyID
@@ -94,6 +97,29 @@ func taskMetrics(task store.GatewayTask, user *auth.User, body map[string]any, c
 	return metrics
 }

+func attemptMetrics(candidate store.RuntimeModelCandidate, attemptNo int, simulated bool) map[string]any {
+	metrics := map[string]any{
+		"resolvedModel":   candidate.ModelName,
+		"modelName":       candidate.ModelName,
+		"modelAlias":      candidate.ModelAlias,
+		"providerModel":   candidate.ProviderModelName,
+		"canonicalModel":  candidate.CanonicalModelKey,
+		"modelType":       candidate.ModelType,
+		"provider":        candidate.Provider,
+		"platformId":      candidate.PlatformID,
+		"platformKey":     candidate.PlatformKey,
+		"platformName":    candidate.PlatformName,
+		"platformModelId": candidate.PlatformModelID,
+		"clientId":        candidate.ClientID,
+		"queueKey":        candidate.QueueKey,
+		"simulated":       simulated,
+	}
+	if attemptNo > 0 {
+		metrics["attempt"] = attemptNo
+	}
+	return metrics
+}
+
 func usageToMap(usage clients.Usage) map[string]any {
 	out := map[string]any{}
 	if usage.InputTokens > 0 {
@@ -172,6 +198,46 @@ func failureMetrics(err error, simulated bool) (string, map[string]any, time.Tim
 	return meta.RequestID, metrics, meta.ResponseStartedAt, meta.ResponseFinishedAt, meta.ResponseDurationMS
 }

+func mergeMetrics(values ...map[string]any) map[string]any {
+	out := map[string]any{}
+	for _, value := range values {
+		for key, item := range value {
+			out[key] = item
+		}
+	}
+	return out
+}
+
+func summarizeAttempts(attempts []store.TaskAttempt) []map[string]any {
+	items := make([]map[string]any, 0, len(attempts))
+	for _, attempt := range attempts {
+		item := map[string]any{
+			"attempt":            attempt.AttemptNo,
+			"status":             attempt.Status,
+			"platformId":         attempt.PlatformID,
+			"platformName":       attempt.PlatformName,
+			"provider":           attempt.Provider,
+			"platformModelId":    attempt.PlatformModelID,
+			"modelName":          attempt.ModelName,
+			"providerModelName":  attempt.ProviderModelName,
+			"modelAlias":         attempt.ModelAlias,
+			"modelType":          attempt.ModelType,
+			"clientId":           attempt.ClientID,
+			"queueKey":           attempt.QueueKey,
+			"requestId":          attempt.RequestID,
+			"retryable":          attempt.Retryable,
+			"simulated":          attempt.Simulated,
+			"errorCode":          attempt.ErrorCode,
+			"errorMessage":       attempt.ErrorMessage,
+			"responseDurationMs": attempt.ResponseDurationMS,
+			"startedAt":          attempt.StartedAt,
+			"finishedAt":         attempt.FinishedAt,
+		}
+		items = append(items, item)
+	}
+	return items
+}
+
 func messageCount(body map[string]any) int {
 	messages, _ := body["messages"].([]any)
 	return len(messages)