wangbo
2bfeb3e179
生产同构 Gemini 千请求在真实候选并发饱和后因 queueing disabled 返回 429，同时单台压测机建立千条 WAN TCP 产生连接层 reset。仅对 acceptance 与 acceptance_canary 覆盖为 10000 条、最长 15 分钟的有界等待队列，保留候选原始并发和 RPM/TPM 上限；压测传输强制 HTTP/2 复用连接，避免单源连接风暴。正式 production 策略不变。\n\n验证：acceptance-load 与 runner 定向测试通过，新增生产策略不变和验收队列边界单测；gofmt、bash -n、ShellCheck、git diff --check 通过。
64 lines
2.2 KiB
Go
64 lines
2.2 KiB
Go
package runner
|
|
|
|
import (
|
|
"testing"
|
|
|
|
"github.com/easyai/easyai-ai-gateway/apps/api/internal/store"
|
|
)
|
|
|
|
func TestDistributedAdmissionModelTypeBoundary(t *testing.T) {
|
|
for _, modelType := range []string{
|
|
"image_generate", "image_edit", "image_analysis", "image_vectorize",
|
|
"video_generate", "video_enhance", "image_to_video", "text_to_video",
|
|
"video_edit", "video_reference", "video_first_last_frame", "video_understanding", "omni_video", "omni",
|
|
"audio_generate", "audio_understanding", "text_to_speech", "voice_clone",
|
|
} {
|
|
if !distributedAdmissionModelType(modelType) {
|
|
t.Errorf("%s should use distributed admission", modelType)
|
|
}
|
|
}
|
|
for _, modelType := range []string{"text_generate", "embedding", "rerank", "", "unknown"} {
|
|
if distributedAdmissionModelType(modelType) {
|
|
t.Errorf("%s should preserve immediate execution/rate-limit behavior", modelType)
|
|
}
|
|
}
|
|
}
|
|
|
|
func TestAcceptanceAdmissionScopesEnableBoundedQueueWithoutChangingConcurrency(t *testing.T) {
|
|
input := []store.AdmissionScope{{
|
|
ScopeType: "platform_model",
|
|
ScopeKey: "model-1",
|
|
ConcurrentLimit: 10,
|
|
QueueLimit: 0,
|
|
MaxWaitSeconds: 0,
|
|
}}
|
|
|
|
got := acceptanceAdmissionScopes(store.GatewayTask{RunMode: "acceptance"}, input)
|
|
|
|
if got[0].ConcurrentLimit != 10 {
|
|
t.Fatalf("acceptance must preserve the production concurrency limit, got %+v", got[0])
|
|
}
|
|
if got[0].QueueLimit != acceptanceQueueLimit || got[0].MaxWaitSeconds != acceptanceQueueMaxWait {
|
|
t.Fatalf("acceptance must enable a bounded queue, got %+v", got[0])
|
|
}
|
|
if input[0].QueueLimit != 0 || input[0].MaxWaitSeconds != 0 {
|
|
t.Fatalf("acceptance queue overlay must not mutate the production scopes, got %+v", input[0])
|
|
}
|
|
}
|
|
|
|
func TestAcceptanceAdmissionScopesLeaveProductionPolicyUnchanged(t *testing.T) {
|
|
input := []store.AdmissionScope{{
|
|
ScopeType: "platform_model",
|
|
ScopeKey: "model-1",
|
|
ConcurrentLimit: 10,
|
|
QueueLimit: 3,
|
|
MaxWaitSeconds: 7,
|
|
}}
|
|
|
|
got := acceptanceAdmissionScopes(store.GatewayTask{RunMode: "production"}, input)
|
|
|
|
if got[0].QueueLimit != 3 || got[0].MaxWaitSeconds != 7 || got[0].ConcurrentLimit != 10 {
|
|
t.Fatalf("production admission policy changed: %+v", got[0])
|
|
}
|
|
}
|