fix(acceptance): 按隔离额度执行平台降级路由

让 acceptance 候选负载读取同一 Run ID 下的并发、RPM 和 TPM 计数器,并在单个平台模型额度不足时继续尝试下一候选,生产与 canary 额度作用域保持不变。\n\n补齐本地容量控制器对宁波和香港测试 Worker 的最小 RBAC,避免初次应用生产反亲和配置时发生滚动死锁,并将突发负载对齐到完整限流窗口。\n\n验证:\n- go test ./internal/runner ./internal/store -count=1\n- go vet ./internal/runner ./internal/store\n- bash -n scripts/acceptance/local-cluster.sh scripts/acceptance/provider-burst.sh scripts/acceptance/run-local-acceptance.sh\n- shellcheck -x scripts/acceptance/local-cluster.sh scripts/acceptance/provider-burst.sh scripts/acceptance/run-local-acceptance.sh\n- kubectl apply --dry-run=client -f deploy/kubernetes/local-acceptance/capacity-controller-rbac.yaml
This commit is contained in:
2026-08-03 12:06:09 +08:00
parent 8485f14bc4
commit d740e9f676
7 changed files with 67 additions and 4 deletions
+1
View File
@@ -372,6 +372,7 @@ func (s *Service) executeWithToken(ctx context.Context, task store.GatewayTask,
CacheAffinityKey: cacheAffinityKeys.Primary,
CacheAffinityKeys: cacheAffinityKeys.Lookup,
CacheAffinityPolicy: runnerPolicy.CacheAffinityPolicy,
QuotaScopePrefix: acceptanceQuotaScopePrefix(task),
})
if err == nil {
candidates, err = filterCandidatesByRequestedPlatform(candidates, body)