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 TestAcceptanceAdmissionScopesUseWorkerCapacityInsteadOfSupplierConcurrency(t *testing.T) { input := []store.AdmissionScope{{ ScopeType: "platform_model", ScopeKey: "model-1", ConcurrentLimit: 10, QueueLimit: 0, MaxWaitSeconds: 0, }, { ScopeType: "worker_capacity", ScopeKey: "global", ConcurrentLimit: 48, }} got := acceptanceAdmissionScopes(store.GatewayTask{RunMode: "acceptance"}, input) if got[0].ConcurrentLimit != 0 { t.Fatalf("protocol-emulated acceptance must defer to worker capacity, 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 got[1].ConcurrentLimit != 48 { t.Fatalf("acceptance worker capacity changed, got %+v", got[1]) } 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 TestAcceptanceCanaryPreservesProductionConcurrency(t *testing.T) { input := []store.AdmissionScope{{ ScopeType: "platform_model", ScopeKey: "model-1", ConcurrentLimit: 10, }} got := acceptanceAdmissionScopes(store.GatewayTask{RunMode: "acceptance_canary"}, input) if got[0].ConcurrentLimit != 10 { t.Fatalf("real acceptance canary must preserve production concurrency, got %+v", got[0]) } } func TestAcceptanceInfrastructureReservationsOnlyRemoveSupplierConcurrency(t *testing.T) { input := []store.RateLimitReservation{ {ScopeType: "platform_model", Metric: "concurrent", Limit: 10}, {ScopeType: "platform_model", Metric: "rpm", Limit: 600}, {ScopeType: "user_group", Metric: "concurrent", Limit: 20}, } got := acceptanceInfrastructureReservations(store.GatewayTask{RunMode: "acceptance"}, input) if len(got) != 2 || got[0].Metric != "rpm" || got[1].ScopeType != "user_group" { t.Fatalf("unexpected acceptance reservations: %+v", got) } canary := acceptanceInfrastructureReservations(store.GatewayTask{RunMode: "acceptance_canary"}, input) if len(canary) != len(input) { t.Fatalf("real canary reservations changed: %+v", canary) } } 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]) } } func TestAsyncAdmissionBatchCoversCertifiedGlobalCapacity(t *testing.T) { for _, testCase := range []struct { globalHardLimit int want int }{ {globalHardLimit: 48, want: 48}, {globalHardLimit: 64, want: 64}, {globalHardLimit: 128, want: 64}, {globalHardLimit: 0, want: 64}, } { if got := asyncAdmissionBatchLimit(testCase.globalHardLimit); got != testCase.want { t.Fatalf("async admission batch limit for %d = %d, want %d", testCase.globalHardLimit, got, testCase.want) } } } func TestPinCandidatesToTaskAdmissionPreservesAdmittedCandidate(t *testing.T) { input := []store.RuntimeModelCandidate{ {PlatformID: "platform-a", PlatformModelID: "model-a"}, {PlatformID: "platform-b", PlatformModelID: "model-b"}, {PlatformID: "platform-c", PlatformModelID: "model-c"}, } admission := &store.TaskAdmission{ Status: "admitted", PlatformID: "platform-b", PlatformModelID: "model-b", } got, pinned := pinCandidatesToTaskAdmission(input, admission) if !pinned { t.Fatal("expected admitted candidate to be pinned") } if got[0].PlatformModelID != "model-b" || got[1].PlatformModelID != "model-a" || got[2].PlatformModelID != "model-c" { t.Fatalf("unexpected candidate order: %+v", got) } if input[0].PlatformModelID != "model-a" { t.Fatalf("candidate pinning mutated the caller slice: %+v", input) } } func TestPinCandidatesToTaskAdmissionPreservesWaitingCandidate(t *testing.T) { input := []store.RuntimeModelCandidate{ {PlatformID: "platform-a", PlatformModelID: "model-a"}, {PlatformID: "platform-b", PlatformModelID: "model-b"}, } admission := &store.TaskAdmission{ Status: "waiting", PlatformID: "platform-b", PlatformModelID: "model-b", } got, pinned := pinCandidatesToTaskAdmission(input, admission) if !pinned || got[0].PlatformModelID != "model-b" { t.Fatalf("waiting candidate was not pinned: pinned=%v candidates=%+v", pinned, got) } } func TestPinCandidatesToTaskAdmissionIgnoresMissingCandidate(t *testing.T) { input := []store.RuntimeModelCandidate{ {PlatformID: "platform-a", PlatformModelID: "model-a"}, {PlatformID: "platform-b", PlatformModelID: "model-b"}, } admission := &store.TaskAdmission{ Status: "admitted", PlatformID: "platform-c", PlatformModelID: "model-c", } got, pinned := pinCandidatesToTaskAdmission(input, admission) if pinned { t.Fatalf("unexpected candidate pin for missing binding: %+v", got) } if got[0].PlatformModelID != "model-a" { t.Fatalf("candidate order changed for missing binding: %+v", got) } }