hobokenchicken
1f574d8134
- Circuit breaker: proper thresholds (3 failures, 30s timeout) - HTTP timeouts: 30s on all providers (was no timeout) - Structured logging: slog replaces fmt.Printf throughout - Stream errors: propagated as SSE error events to client - Registry fetch: retry with backoff (3 attempts) - Registry reads in dashboard protected by RWMutex
77 lines
1.9 KiB
Go
77 lines
1.9 KiB
Go
package utils
|
|
|
|
import (
|
|
"encoding/json"
|
|
"fmt"
|
|
"log"
|
|
"time"
|
|
|
|
"github.com/go-resty/resty/v2"
|
|
"gophergate/internal/models"
|
|
)
|
|
|
|
const ModelsDevURL = "https://models.dev/api.json"
|
|
|
|
func FetchRegistry() (*models.ModelRegistry, error) {
|
|
client := resty.New().SetTimeout(10 * time.Second)
|
|
|
|
var lastErr error
|
|
for attempt := 0; attempt < 3; attempt++ {
|
|
if attempt > 0 {
|
|
backoff := time.Duration(1<<attempt) * time.Second
|
|
time.Sleep(backoff)
|
|
}
|
|
|
|
resp, err := client.R().Get(ModelsDevURL)
|
|
if err != nil {
|
|
lastErr = fmt.Errorf("attempt %d: %w", attempt+1, err)
|
|
continue
|
|
}
|
|
if !resp.IsSuccess() {
|
|
lastErr = fmt.Errorf("attempt %d: HTTP %d", attempt+1, resp.StatusCode())
|
|
continue
|
|
}
|
|
|
|
var providers map[string]models.ProviderInfo
|
|
if err := json.Unmarshal(resp.Body(), &providers); err != nil {
|
|
lastErr = fmt.Errorf("attempt %d: unmarshal: %w", attempt+1, err)
|
|
continue
|
|
}
|
|
|
|
log.Println("Successfully loaded model registry")
|
|
return &models.ModelRegistry{Providers: providers}, nil
|
|
}
|
|
|
|
return nil, fmt.Errorf("failed to fetch registry after 3 attempts: %w", lastErr)
|
|
}
|
|
|
|
func CalculateCost(registry *models.ModelRegistry, modelID string, promptTokens, completionTokens, reasoningTokens, cacheRead, cacheWrite uint32) float64 {
|
|
meta := registry.FindModel(modelID)
|
|
if meta == nil || meta.Cost == nil {
|
|
return 0.0
|
|
}
|
|
|
|
// promptTokens is usually the TOTAL prompt size.
|
|
// We subtract cacheRead from it to get the uncached part.
|
|
uncachedTokens := promptTokens
|
|
if cacheRead > 0 {
|
|
if cacheRead > promptTokens {
|
|
uncachedTokens = 0
|
|
} else {
|
|
uncachedTokens = promptTokens - cacheRead
|
|
}
|
|
}
|
|
|
|
cost := (float64(uncachedTokens) * meta.Cost.Input / 1000000.0) +
|
|
(float64(completionTokens) * meta.Cost.Output / 1000000.0)
|
|
|
|
if meta.Cost.CacheRead != nil {
|
|
cost += float64(cacheRead) * (*meta.Cost.CacheRead) / 1000000.0
|
|
}
|
|
if meta.Cost.CacheWrite != nil {
|
|
cost += float64(cacheWrite) * (*meta.Cost.CacheWrite) / 1000000.0
|
|
}
|
|
|
|
return cost
|
|
}
|