LLM/CLIProxyAPIPlus
1
0
Fork 0
mirror of https://github.com/router-for-me/CLIProxyAPIPlus.git synced 2026-04-13 01:34:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(auth): handle OAuth model alias in retry logic and refine Qwen quota handling

Browse Source
This commit is contained in:
Luis Pater
2026-04-09 03:44:19 +08:00
parent d54f816363
commit 941334da79
4 changed files with 76 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
internal/runtime/executor/qwen_executor.go
View File

@@ -32,16 +32,6 @@ const (
var qwenDefaultSystemMessage = []byte(`{"role":"system","content":[{"type":"text","text":"","cache_control":{"type":"ephemeral"}}]}`)
// qwenBeijingLoc caches the Beijing timezone to avoid repeated LoadLocation syscalls.
var qwenBeijingLoc = func() *time.Location {
loc, err := time.LoadLocation("Asia/Shanghai")
if err != nil || loc == nil {
log.Warnf("qwen: failed to load Asia/Shanghai timezone: %v, using fixed UTC+8", err)
return time.FixedZone("CST", 8*3600)
}
return loc
}()
// qwenQuotaCodes is a package-level set of error codes that indicate quota exhaustion.
var qwenQuotaCodes = map[string]struct{}{
"insufficient_quota": {},
@@ -156,22 +146,13 @@ func wrapQwenError(ctx context.Context, httpCode int, body []byte) (errCode int,
// Qwen returns 403 for quota errors, 429 for rate limits
if (httpCode == http.StatusForbidden || httpCode == http.StatusTooManyRequests) && isQwenQuotaError(body) {
errCode = http.StatusTooManyRequests // Map to 429 to trigger quota logic
cooldown := timeUntilNextDay()
retryAfter = &cooldown
helps.LogWithRequestID(ctx).Warnf("qwen quota exceeded (http %d -> %d), cooling down until tomorrow (%v)", httpCode, errCode, cooldown)
// Do not force an excessively long retry-after (e.g. until tomorrow), otherwise
// the global request-retry scheduler may skip retries due to max-retry-interval.
helps.LogWithRequestID(ctx).Warnf("qwen quota exceeded (http %d -> %d)", httpCode, errCode)
}
return errCode, retryAfter
}
// timeUntilNextDay returns duration until midnight Beijing time (UTC+8).
// Qwen's daily quota resets at 00:00 Beijing time.
func timeUntilNextDay() time.Duration {
now := time.Now()
nowLocal := now.In(qwenBeijingLoc)
tomorrow := time.Date(nowLocal.Year(), nowLocal.Month(), nowLocal.Day()+1, 0, 0, 0, 0, qwenBeijingLoc)
return tomorrow.Sub(now)
}
// ensureQwenSystemMessage ensures the request has a single system message at the beginning.
// It always injects the default system prompt and merges any user-provided system messages
// into the injected system message content to satisfy Qwen's strict message ordering rules.

24
internal/runtime/executor/qwen_executor_test.go
View File

@@ -1,6 +1,8 @@
package executor
import (
"context"
"net/http"
"testing"
"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
@@ -152,3 +154,25 @@ func TestEnsureQwenSystemMessage_MergesMultipleSystemMessages(t *testing.T) {
t.Fatalf("messages[0].content[2].text = %q, want %q", parts[2].Get("text").String(), "B")
}
}
func TestWrapQwenError_InsufficientQuotaDoesNotSetRetryAfter(t *testing.T) {
body := []byte(`{"error":{"code":"insufficient_quota","message":"You exceeded your current quota","type":"insufficient_quota"}}`)
code, retryAfter := wrapQwenError(context.Background(), http.StatusTooManyRequests, body)
if code != http.StatusTooManyRequests {
t.Fatalf("wrapQwenError status = %d, want %d", code, http.StatusTooManyRequests)
}
if retryAfter != nil {
t.Fatalf("wrapQwenError retryAfter = %v, want nil", *retryAfter)
}
}
func TestWrapQwenError_Maps403QuotaTo429WithoutRetryAfter(t *testing.T) {
body := []byte(`{"error":{"code":"insufficient_quota","message":"You exceeded your current quota","type":"insufficient_quota"}}`)
code, retryAfter := wrapQwenError(context.Background(), http.StatusForbidden, body)
if code != http.StatusTooManyRequests {
t.Fatalf("wrapQwenError status = %d, want %d", code, http.StatusTooManyRequests)
}
if retryAfter != nil {
t.Fatalf("wrapQwenError retryAfter = %v, want nil", *retryAfter)
}
}