test(acceptance): 验证厂商三类额度与优先路由

为 provider-burst 固定三平台并发、RPM、TPM 和优先级组合，使用稳定 token 用量的本地视频负载验证同级按容量分流及低优先级溢出。\n\n报告升级为 v2，强制校验三类额度峰值精确命中上限、不越界、队列回收、三 Worker 分发以及重复提交和回调安全。\n\n验证：\n- env -u AI_GATEWAY_TEST_DATABASE_URL go test ./... -count=1\n- go vet ./...\n- bash -n scripts/acceptance/provider-burst.sh scripts/acceptance/run-local-acceptance.sh\n- shellcheck -x scripts/acceptance/provider-burst.sh scripts/acceptance/run-local-acceptance.sh

apps/api/cmd/acceptance-load/main.go

@@ -279,7 +279,7 @@ func parseOptions() (options, error) {
 		return options{}, errors.New("requests must be between 0 and 10000")
 	}
 	switch opts.profile {
-	case "simulated-smoke", "simulated-all", "gemini-baseline", "gemini-multi-image", "gemini-large", "gemini-peak", "video-throughput", "video-capacity", "video-recovery",
+	case "simulated-smoke", "simulated-all", "gemini-baseline", "gemini-multi-image", "gemini-large", "gemini-peak", "video-throughput", "video-capacity", "video-provider-quota", "video-recovery",
 		"mixed-soak", "mixed-overload":
 		if opts.emulatorURL == "" {
 			return options{}, errors.New("AI_GATEWAY_ACCEPTANCE_EMULATOR_URL is required for simulated profiles")
@@ -371,6 +371,12 @@ func run(ctx context.Context, opts options) ([]phaseResult, error) {
 					requests = opts.requestCount
 				}
 				result = runVideo(ctx, client, opts, name, opts.shardRequestCount(requests), false, opts.emulatorFixtureURLs(), false, opts.shardIndex, opts.shardCount)
+			case "video-provider-quota":
+				requests := 24
+				if opts.requestCount > 0 {
+					requests = opts.requestCount
+				}
+				result = runVideo(ctx, client, opts, name, opts.shardRequestCount(requests), false, opts.emulatorFixtureURLs(), false, opts.shardIndex, opts.shardCount)
 			case "video-recovery":
 				result = runVideo(ctx, client, opts, name, opts.shardRequestCount(96), true, opts.emulatorFixtureURLs(), false, opts.shardIndex, opts.shardCount)
 			case "mixed-soak", "mixed-overload":
@@ -657,18 +663,21 @@ func runVideo(
 			if len(combo) > imageCount {
 				combo = combo[:imageCount]
 			}
-			if !realUpstream && logicalIndex%4 == 0 {
+			if !realUpstream && name != "video-provider-quota" && logicalIndex%4 == 0 {
 				combo[0] = imageURLs[12+(logicalIndex/4)%4]
 			}
 			content := make([]any, 0, len(combo)+1)
 			prompt := "多参考图生成连续运镜视频，保持人物、服装和场景一致"
+			if name == "video-provider-quota" {
+				prompt = "acceptance-provider-quota"
+			}
 			if longRun {
 				prompt += " acceptance-long-recovery"
 			}
 			if name == "video-capacity" {
 				prompt += " acceptance-capacity-ladder"
 			}
-			if !realUpstream && logicalIndex%4 == 0 {
+			if !realUpstream && name != "video-provider-quota" && logicalIndex%4 == 0 {
 				prompt += " acceptance-force-conversion"
 			}
 			content = append(content, map[string]any{"type": "text", "text": prompt})
@@ -716,7 +725,7 @@ func runVideo(
 			err = withOperation("video_submit", err)
 			mu.Lock()
 			latencies[index] = time.Since(requestStarted)
-			if !realUpstream && logicalIndex%4 == 0 {
+			if !realUpstream && name != "video-provider-quota" && logicalIndex%4 == 0 {
 				forcedConversions++
 			}
 			if err != nil {

apps/api/internal/acceptanceemulator/server.go

@@ -78,11 +78,12 @@ type Report struct {
 }
 
 type videoTask struct {
-	ID        string
+	ID          string
-	Model     string
+	Model       string
-	CreatedAt time.Time
+	CreatedAt   time.Time
-	ReadyAt   time.Time
+	ReadyAt     time.Time
-	ImageRefs []imageReference
+	UsageTokens int
+	ImageRefs   []imageReference
 }
 
 type imageReference struct {
@@ -243,7 +244,7 @@ func (s *Server) submitVideo(w http.ResponseWriter, r *http.Request) {
 	id := "acceptance-video-" + strconv.FormatUint(s.nextID.Add(1), 10)
 	task := videoTask{
 		ID: id, Model: strings.TrimSpace(stringValue(body["model"])),
-		CreatedAt: s.now(), ReadyAt: s.now().Add(delay), ImageRefs: refs,
+		CreatedAt: s.now(), ReadyAt: s.now().Add(delay), UsageTokens: videoUsageTokens(body), ImageRefs: refs,
 	}
 	s.mu.Lock()
 	if existingID := s.videoByIdempotency[idempotencyKey]; existingID != "" {
@@ -298,7 +299,7 @@ func (s *Server) getVideo(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusOK, map[string]any{
 		"id": id, "model": task.Model, "status": status, "content": content,
 		"created_at": task.CreatedAt.Unix(),
-		"usage":      map[string]any{"completion_tokens": 1, "total_tokens": 1},
+		"usage":      map[string]any{"completion_tokens": task.UsageTokens, "total_tokens": task.UsageTokens},
 	})
 }
 
@@ -674,12 +675,22 @@ func videoDelay(body map[string]any, longRun bool) time.Duration {
 	if longRun {
 		return 2*time.Minute + time.Duration(seed%61)*time.Second
 	}
+	if videoPromptContains(body, "acceptance-provider-quota") {
+		return 5*time.Second + time.Duration(seed%6)*time.Second
+	}
 	if videoPromptContains(body, "acceptance-capacity-ladder") {
 		return 30*time.Second + time.Duration(seed%16)*time.Second
 	}
 	return 5*time.Second + time.Duration(seed%11)*time.Second
 }
 
+func videoUsageTokens(body map[string]any) int {
+	if videoPromptContains(body, "acceptance-provider-quota") {
+		return 7
+	}
+	return 1
+}
+
 func videoPromptContains(body map[string]any, marker string) bool {
 	content, _ := body["content"].([]any)
 	for _, raw := range content {

apps/api/internal/acceptanceemulator/server_test.go

@@ -242,6 +242,23 @@ func TestVideoCapacityDelay(t *testing.T) {
 	}
 }
 
+func TestVideoProviderQuotaDelayAndUsage(t *testing.T) {
+	body := map[string]any{
+		"seed": json.Number("5"),
+		"content": []any{map[string]any{
+			"type": "text",
+			"text": "acceptance-provider-quota",
+		}},
+	}
+	delay := videoDelay(body, false)
+	if delay < 5*time.Second || delay > 10*time.Second {
+		t.Fatalf("video provider quota delay=%s, want 5s..10s", delay)
+	}
+	if tokens := videoUsageTokens(body); tokens != 7 {
+		t.Fatalf("video provider quota usage tokens=%d, want 7", tokens)
+	}
+}
+
 func TestVolcesProtocolAcceptsThreeSixAndNineReferenceImages(t *testing.T) {
 	server := New(Config{Wait: func(context.Context, time.Duration) error { return nil }})
 	httpServer := httptest.NewServer(server.Handler())

apps/api/internal/runner/limits_test.go

@@ -134,3 +134,15 @@ func TestReservationsFromPolicySkipsNonPositiveLimits(t *testing.T) {
 		t.Fatalf("expected concurrent reservation with limit 2, got %+v", reservations[0])
 	}
 }
+
+func TestProviderQuotaPromptReservesSevenTokens(t *testing.T) {
+	body := map[string]any{
+		"content": []any{map[string]any{
+			"type": "text",
+			"text": "acceptance-provider-quota",
+		}},
+	}
+	if got := estimateRequestTokens(body); got != 7 {
+		t.Fatalf("provider quota prompt tokens=%d, want 7", got)
+	}
+}

scripts/acceptance/provider-burst.sh

@@ -44,11 +44,11 @@ WITH source AS (
     AND COALESCE((platform.config->>'acceptanceSnapshot')::boolean, false)
   ORDER BY platform.priority, platform.created_at
   LIMIT 1
-), variants(platform_id, platform_key, platform_name, concurrency_limit) AS (
+), variants(platform_id, platform_key, platform_name, concurrency_limit, rpm_limit, tpm_limit, priority) AS (
   VALUES
-    ('${provider_burst_platform_ids[0]}'::uuid, 'acceptance-provider-burst-2', 'Acceptance Provider Burst C2', 2),
+    ('${provider_burst_platform_ids[0]}'::uuid, 'acceptance-provider-burst-2', 'Acceptance Provider Burst C2', 2, 4, 28, 100),
-    ('${provider_burst_platform_ids[1]}'::uuid, 'acceptance-provider-burst-4', 'Acceptance Provider Burst C4', 4),
+    ('${provider_burst_platform_ids[1]}'::uuid, 'acceptance-provider-burst-4', 'Acceptance Provider Burst C4', 4, 8, 56, 100),
-    ('${provider_burst_platform_ids[2]}'::uuid, 'acceptance-provider-burst-6', 'Acceptance Provider Burst C6', 6)
+    ('${provider_burst_platform_ids[2]}'::uuid, 'acceptance-provider-burst-6', 'Acceptance Provider Burst C6', 6, 12, 84, 200)
 )
 INSERT INTO integration_platforms (
   id, provider, platform_key, name, base_url, auth_type, credentials, config,
@@ -61,12 +61,14 @@ SELECT variants.platform_id, source.provider, variants.platform_key, variants.pl
        source.config || jsonb_build_object(
          'acceptanceProviderBurst', true,
          'acceptanceBurstRunId', '$run_id',
-         'acceptanceConcurrencyLimit', variants.concurrency_limit
+         'acceptanceConcurrencyLimit', variants.concurrency_limit,
+         'acceptanceRPMLimit', variants.rpm_limit,
+         'acceptanceTPMLimit', variants.tpm_limit
        ),
        source.visibility_scope, source.tenant_id, source.tenant_key,
        source.default_pricing_mode, source.default_discount_factor,
        source.pricing_rule_set_id, source.retry_policy, '{\"rules\":[]}'::jsonb,
-       100, NULL, 'enabled', NULL, NULL, variants.platform_name
+       variants.priority, NULL, 'enabled', NULL, NULL, variants.platform_name
 FROM source CROSS JOIN variants
 ON CONFLICT (id) DO UPDATE SET
   provider = EXCLUDED.provider,
@@ -102,11 +104,11 @@ WITH source AS (
     AND COALESCE((platform.config->>'acceptanceSnapshot')::boolean, false)
   ORDER BY platform.priority, platform.created_at
   LIMIT 1
-), variants(model_id, platform_id, concurrency_limit) AS (
+), variants(model_id, platform_id, concurrency_limit, rpm_limit, tpm_limit) AS (
   VALUES
-    ('${provider_burst_model_ids[0]}'::uuid, '${provider_burst_platform_ids[0]}'::uuid, 2),
+    ('${provider_burst_model_ids[0]}'::uuid, '${provider_burst_platform_ids[0]}'::uuid, 2, 4, 28),
-    ('${provider_burst_model_ids[1]}'::uuid, '${provider_burst_platform_ids[1]}'::uuid, 4),
+    ('${provider_burst_model_ids[1]}'::uuid, '${provider_burst_platform_ids[1]}'::uuid, 4, 8, 56),
-    ('${provider_burst_model_ids[2]}'::uuid, '${provider_burst_platform_ids[2]}'::uuid, 6)
+    ('${provider_burst_model_ids[2]}'::uuid, '${provider_burst_platform_ids[2]}'::uuid, 6, 12, 84)
 )
 INSERT INTO platform_models (
   id, platform_id, base_model_id, model_name, model_alias, model_type,
@@ -127,7 +129,8 @@ SELECT variants.model_id, variants.platform_id, source.base_model_id,
            'metric', 'concurrent', 'limit', variants.concurrency_limit,
            'leaseTtlSeconds', 120
          ),
-         jsonb_build_object('metric', 'rpm', 'limit', 600, 'windowSeconds', 60)
+         jsonb_build_object('metric', 'rpm', 'limit', variants.rpm_limit, 'windowSeconds', 60),
+         jsonb_build_object('metric', 'tpm_total', 'limit', variants.tpm_limit, 'windowSeconds', 60)
        )),
        'override', NULL, '{}'::jsonb, source.provider_model_name, NULL, true
 FROM source CROSS JOIN variants
@@ -219,13 +222,18 @@ DO UPDATE SET status = 'active', metadata = EXCLUDED.metadata,
 SELECT jsonb_build_object(
   'platforms', count(*),
   'limits', jsonb_agg((platform.config->>'acceptanceConcurrencyLimit')::int ORDER BY (platform.config->>'acceptanceConcurrencyLimit')::int),
+  'rpmLimits', jsonb_agg((platform.config->>'acceptanceRPMLimit')::int ORDER BY (platform.config->>'acceptanceConcurrencyLimit')::int),
+  'tpmLimits', jsonb_agg((platform.config->>'acceptanceTPMLimit')::int ORDER BY (platform.config->>'acceptanceConcurrencyLimit')::int),
+  'priorities', jsonb_agg(platform.priority ORDER BY (platform.config->>'acceptanceConcurrencyLimit')::int),
   'enabled', count(*) FILTER (WHERE platform.status='enabled' AND model.enabled)
 )
 FROM integration_platforms platform
 JOIN platform_models model ON model.platform_id = platform.id
 WHERE COALESCE((platform.config->>'acceptanceProviderBurst')::boolean, false)
   AND platform.config->>'acceptanceBurstRunId' = '$run_id';")
-  jq -e '.platforms == 3 and .enabled == 3 and .limits == [2,4,6]' <<<"$configured" >/dev/null
+  jq -e '.platforms == 3 and .enabled == 3 and .limits == [2,4,6] and
+    .rpmLimits == [4,8,12] and .tpmLimits == [28,56,84] and .priorities == [100,100,200]' \
+    <<<"$configured" >/dev/null
 }
 
 sample_provider_burst() {
@@ -235,7 +243,10 @@ sample_provider_burst() {
     database_query "
 WITH configured AS (
   SELECT platform.id platform_id, platform.platform_key, model.id platform_model_id,
-         (platform.config->>'acceptanceConcurrencyLimit')::int concurrency_limit
+         platform.priority,
+         (platform.config->>'acceptanceConcurrencyLimit')::int concurrency_limit,
+         (platform.config->>'acceptanceRPMLimit')::int rpm_limit,
+         (platform.config->>'acceptanceTPMLimit')::int tpm_limit
   FROM integration_platforms platform
   JOIN platform_models model ON model.platform_id = platform.id
   WHERE COALESCE((platform.config->>'acceptanceProviderBurst')::boolean, false)
@@ -302,7 +313,10 @@ build_provider_burst_report() {
   platforms=$(database_query "
 WITH configured AS (
   SELECT platform.id platform_id, platform.platform_key, model.id platform_model_id,
-         (platform.config->>'acceptanceConcurrencyLimit')::int concurrency_limit
+         platform.priority,
+         (platform.config->>'acceptanceConcurrencyLimit')::int concurrency_limit,
+         (platform.config->>'acceptanceRPMLimit')::int rpm_limit,
+         (platform.config->>'acceptanceTPMLimit')::int tpm_limit
   FROM integration_platforms platform
   JOIN platform_models model ON model.platform_id=platform.id
   WHERE COALESCE((platform.config->>'acceptanceProviderBurst')::boolean, false)
@@ -332,18 +346,34 @@ WITH configured AS (
   FROM gateway_task_attempts attempt
   WHERE attempt.task_id IN (SELECT id FROM gateway_tasks WHERE acceptance_run_id='$run_id'::uuid)
   GROUP BY attempt.platform_model_id
+), rate_peaks AS (
+  SELECT split_part(counter.scope_key, ':', 3)::uuid platform_model_id,
+         max(counter.limit_value) FILTER (WHERE counter.metric='rpm') rpm_limit,
+         max(counter.used_value + counter.reserved_value) FILTER (WHERE counter.metric='rpm') rpm_peak,
+         max(counter.limit_value) FILTER (WHERE counter.metric='tpm_total') tpm_limit,
+         max(counter.used_value + counter.reserved_value) FILTER (WHERE counter.metric='tpm_total') tpm_peak
+  FROM gateway_rate_limit_counters counter
+  WHERE counter.scope_type='platform_model'
+    AND counter.scope_key LIKE 'acceptance:$run_id:%'
+  GROUP BY split_part(counter.scope_key, ':', 3)::uuid
 )
 SELECT COALESCE(jsonb_agg(jsonb_build_object(
   'platformId', configured.platform_id,
   'platformModelId', configured.platform_model_id,
   'platformKey', configured.platform_key,
+  'priority', configured.priority,
   'limit', configured.concurrency_limit,
   'peak', COALESCE(peaks.peak,0),
+  'rpmLimit', configured.rpm_limit,
+  'rpmPeak', COALESCE(rate_peaks.rpm_peak,0),
+  'tpmLimit', configured.tpm_limit,
+  'tpmPeak', COALESCE(rate_peaks.tpm_peak,0),
   'attempts', COALESCE(attempts.attempts,0)
 ) ORDER BY configured.concurrency_limit), '[]'::jsonb)
 FROM configured
 LEFT JOIN peaks ON peaks.platform_model_id=configured.platform_model_id
-LEFT JOIN attempts ON attempts.platform_model_id=configured.platform_model_id;")
+LEFT JOIN attempts ON attempts.platform_model_id=configured.platform_model_id
+LEFT JOIN rate_peaks ON rate_peaks.platform_model_id=configured.platform_model_id;")
   queue=$(jq -s '{
     samples:length,
     maxQueuedTasks:(map(.queuedTasks)|max),
@@ -437,7 +467,16 @@ SELECT jsonb_build_object(
 FROM gateway_task_callback_outbox callback
 WHERE callback.task_id IN (SELECT id FROM gateway_tasks WHERE acceptance_run_id='$run_id'::uuid);")
   resource_summary=$(provider_burst_resource_summary "$resources")
-  if ! jq -e 'length==3 and all(.[]; .peak == .limit and .attempts > 0)' <<<"$platforms" >/dev/null; then passed=false; fi
+  if ! jq -e 'length==3 and
+    [.[].priority] == [100,100,200] and
+    [.[].attempts] == [4,8,12] and
+    all(.[]; .peak == .limit and .rpmPeak == .rpmLimit and .tpmPeak == .tpmLimit)' \
+    <<<"$platforms" >/dev/null; then passed=false; fi
+  if ! jq -e '.[0].priority==.[1].priority and
+    .[0].attempts*2==.[1].attempts and
+    .[2].priority>.[1].priority and
+    .[0].peak==.[0].limit and .[1].peak==.[1].limit and .[2].attempts>0' \
+    <<<"$platforms" >/dev/null; then passed=false; fi
   if ! jq -e '.queueObserved and .admissionWaitObserved and .drainedAtEnd and .maxQueuedTasks > 0 and .maxWaitingAdmissions > 0' <<<"$queue" >/dev/null; then passed=false; fi
   if ! jq -e --argjson requests "$requests" 'length==3 and all(.[]; .tasks>0 and .workerInstanceId!="unclaimed") and ([.[].tasks]|add)==$requests' <<<"$distribution" >/dev/null; then passed=false; fi
   if ! jq -e 'length==3 and all(.[]; .peakActiveTasks>0 and (.pressureStates|index("critical")|not))' <<<"$worker_peaks" >/dev/null; then passed=false; fi
@@ -458,11 +497,20 @@ WHERE callback.task_id IN (SELECT id FROM gateway_tasks WHERE acceptance_run_id=
     --argjson resources "$resource_summary" --argjson leaks "$leaks" \
     --argjson duplicates "$duplicates" --argjson callbacks "$callbacks" \
     '{
-      schemaVersion:"acceptance-provider-burst-report/v1",
+      schemaVersion:"acceptance-provider-burst-report/v2",
       runId:$runId,profile:"provider-burst",model:$model,
       startedAt:(now|todateiso8601),finishedAt:(now|todateiso8601),
       passed:$passed,secretSafe:true,requests:$requests,phases:$phases,
-      providerRouting:{platforms:$platforms,totalConfiguredConcurrency:([$platforms[].limit]|add)},
+      providerRouting:{
+        platforms:$platforms,
+        totalConfiguredConcurrency:([$platforms[].limit]|add),
+        samePriorityLoadBalanced:($platforms[0].priority==$platforms[1].priority and
+          $platforms[0].attempts*2==$platforms[1].attempts),
+        lowerPrioritySpilloverObserved:($platforms[2].priority>$platforms[1].priority and
+          $platforms[0].peak==$platforms[0].limit and
+          $platforms[1].peak==$platforms[1].limit and
+          $platforms[2].attempts>0)
+      },
       queue:$queue,
       cluster:{workers:$workers,workerPeaks:$workerPeaks,taskDistribution:$distribution,
         rateLimits:$rateLimits,concurrencyPeaks:$concurrencyPeaks,
@@ -473,14 +521,14 @@ WHERE callback.task_id IN (SELECT id FROM gateway_tasks WHERE acceptance_run_id=
 }
 
 provider_burst() {
-  local requests=${AI_GATEWAY_LOCAL_PROVIDER_BURST_REQUESTS:-48}
+  local requests=24
   local load_report="$report_root/provider-burst-load.json"
   local samples="$report_root/provider-burst-samples.ndjson"
   local worker_samples="$report_root/provider-burst-worker-samples.ndjson"
   local resources="$report_root/provider-burst-resources.csv"
   local report="$report_root/provider-burst.json"
   worker_sample_stop="$report_root/provider-burst-samples.stop"
-  [[ $requests =~ ^[0-9]+$ ]] && ((requests >= 24 && requests <= 256))
+  [[ $requests == 24 ]]
   [[ ! -e $load_report && ! -e $samples && ! -e $worker_samples && ! -e $resources && ! -e $report && ! -e $worker_sample_stop ]] ||
     fail_gate acceptance_report_exists "provider burst evidence already exists"
   current_phase=provider_burst_platforms
@@ -495,7 +543,7 @@ provider_burst() {
   sample_resources "$resources" "$worker_sample_stop" &
   active_resource_sampler_pid=$!
   current_phase=provider_burst_load
-  run_load video-throughput "$load_report" -requests "$requests"
+  run_load video-provider-quota "$load_report" -requests "$requests"
   sleep 2
   touch "$worker_sample_stop"
   wait "$active_provider_sampler_pid"