Files
1Panel/agent/utils/ai_tools/vllm/request_metrics.go
T

66 lines
2.1 KiB
Go

package vllm
import (
"encoding/json"
"strconv"
"github.com/1Panel-dev/1Panel/agent/app/model"
"github.com/1Panel-dev/1Panel/agent/utils/re"
)
type requestMetrics struct {
Seconds float64 `json:"seconds"`
PrefixHits *float64 `json:"prefixHits,omitempty"`
PrefixQueries *float64 `json:"prefixQueries,omitempty"`
FinishReasons map[string]float64 `json:"finishReasons,omitempty"`
}
func calculateRequestMetrics(current, previous Metrics, elapsed float64, result *model.MonitorVLLM) error {
sample := requestMetrics{Seconds: elapsed, FinishReasons: map[string]float64{}}
hits := counterDelta(current["prefix_cache_hits_total"], previous["prefix_cache_hits_total"])
queries := counterDelta(current["prefix_cache_queries_total"], previous["prefix_cache_queries_total"])
if hits != nil && queries != nil && *hits <= *queries {
sample.PrefixHits, sample.PrefixQueries = hits, queries
if *queries > 0 {
rate := *hits / *queries * 100
result.PrefixCacheHitRate = &rate
}
}
before := groupFinishReasons(previous["request_success_total"])
for reason, values := range groupFinishReasons(current["request_success_total"]) {
if delta := counterDelta(values, before[reason]); delta != nil {
sample.FinishReasons[reason] = *delta
if result.FinishReasons == nil {
result.FinishReasons = map[string]float64{}
}
result.FinishReasons[reason] = *delta / elapsed
}
}
encoded, err := json.Marshal(sample)
if err != nil {
return err
}
result.RequestDeltas = string(encoded)
return nil
}
func groupFinishReasons(samples map[string]float64) map[string]map[string]float64 {
groups := map[string]map[string]float64{}
for labels, value := range samples {
for _, label := range re.GetRegex(re.VLLMMetricLabelPattern).FindAllStringSubmatch(labels, -1) {
if label[1] != "finished_reason" {
continue
}
reason, err := strconv.Unquote(label[2])
if err != nil || reason == "" {
continue
}
if groups[reason] == nil {
groups[reason] = map[string]float64{}
}
groups[reason][labels] = value
}
}
return groups
}