fix(worker): 固定异步准入候选避免执行槽空转
异步任务在调度器准入后,Worker 会再次按实时负载排序候选;排序变化会让已准入任务退回 waiting,同时继续占用 River 执行槽,导致高并发吞吐塌陷。 执行前复用已持久化的 admitted 候选并稳定置顶,候选失效时仍保留原有重选路径;增加候选固定与非准入场景单测。 验证:Go 全量测试、runner race、go vet、OpenAPI 生成一致性、迁移安全检查通过。
This commit is contained in:
@@ -187,6 +187,45 @@ func acceptanceAdmissionScopes(task store.GatewayTask, scopes []store.AdmissionS
|
||||
return out
|
||||
}
|
||||
|
||||
func (s *Service) loadAsyncTaskAdmission(ctx context.Context, task store.GatewayTask) (*store.TaskAdmission, error) {
|
||||
if !task.AsyncMode {
|
||||
return nil, nil
|
||||
}
|
||||
admission, err := s.store.GetTaskAdmission(ctx, task.ID)
|
||||
if errors.Is(err, pgx.ErrNoRows) {
|
||||
return nil, nil
|
||||
}
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return &admission, nil
|
||||
}
|
||||
|
||||
func pinCandidatesToTaskAdmission(
|
||||
candidates []store.RuntimeModelCandidate,
|
||||
admission *store.TaskAdmission,
|
||||
) ([]store.RuntimeModelCandidate, bool) {
|
||||
if admission == nil || admission.Status != "admitted" || len(candidates) < 2 {
|
||||
return candidates, false
|
||||
}
|
||||
pinnedIndex := -1
|
||||
for index := range candidates {
|
||||
if candidates[index].PlatformID == admission.PlatformID &&
|
||||
candidates[index].PlatformModelID == admission.PlatformModelID {
|
||||
pinnedIndex = index
|
||||
break
|
||||
}
|
||||
}
|
||||
if pinnedIndex <= 0 {
|
||||
return candidates, false
|
||||
}
|
||||
out := append([]store.RuntimeModelCandidate(nil), candidates...)
|
||||
pinned := out[pinnedIndex]
|
||||
copy(out[1:pinnedIndex+1], out[:pinnedIndex])
|
||||
out[0] = pinned
|
||||
return out, true
|
||||
}
|
||||
|
||||
func (s *Service) tryTaskAdmission(ctx context.Context, task store.GatewayTask, plan taskAdmissionPlan, waiterID string) (store.TaskAdmissionResult, error) {
|
||||
return s.tryTaskAdmissionWithAdmittedHook(ctx, task, plan, waiterID, nil)
|
||||
}
|
||||
@@ -195,17 +234,11 @@ func (s *Service) activeAsyncTaskAdmission(
|
||||
ctx context.Context,
|
||||
task store.GatewayTask,
|
||||
plan taskAdmissionPlan,
|
||||
admission *store.TaskAdmission,
|
||||
) (store.TaskAdmissionResult, bool, error) {
|
||||
if !task.AsyncMode {
|
||||
if !task.AsyncMode || admission == nil {
|
||||
return store.TaskAdmissionResult{}, false, nil
|
||||
}
|
||||
admission, err := s.store.GetTaskAdmission(ctx, task.ID)
|
||||
if errors.Is(err, pgx.ErrNoRows) {
|
||||
return store.TaskAdmissionResult{}, false, nil
|
||||
}
|
||||
if err != nil {
|
||||
return store.TaskAdmissionResult{}, false, err
|
||||
}
|
||||
if admission.Status != "admitted" ||
|
||||
admission.PlatformID != plan.Candidate.PlatformID ||
|
||||
admission.PlatformModelID != plan.Candidate.PlatformModelID ||
|
||||
@@ -220,7 +253,7 @@ func (s *Service) activeAsyncTaskAdmission(
|
||||
return store.TaskAdmissionResult{}, false, nil
|
||||
}
|
||||
return store.TaskAdmissionResult{
|
||||
Admission: admission,
|
||||
Admission: *admission,
|
||||
Admitted: true,
|
||||
Leases: leases,
|
||||
}, true, nil
|
||||
|
||||
Reference in New Issue
Block a user