feat(sense): implement Control API v1 [T-011]

2026-08-07 20:54:41 +08:00
parent 0d3d7a22ed
commit a0d239811f
45 changed files with 4844 additions and 83 deletions
@@ -2,6 +2,7 @@ package store

 import (
 	"context"
+	"crypto/sha256"
 	"database/sql"
 	"errors"
 	"fmt"
@@ -286,12 +287,12 @@ func TestPostgresOpenRejectsOverprivilegedRuntimeRole(t *testing.T) {
 	_, admin := openPostgresTestStore(t)
 	ctx := context.Background()
 	if _, err := admin.ExecContext(ctx,
-		`GRANT UPDATE ON bell.site_quota_v1 TO yovision_t010_sense`); err != nil {
+		`GRANT UPDATE ON bell.site_quota_v1 TO yovision_t011_sense`); err != nil {
 		t.Fatal(err)
 	}
 	defer func() {
 		_, _ = admin.ExecContext(context.Background(),
-			`REVOKE UPDATE ON bell.site_quota_v1 FROM yovision_t010_sense`)
+			`REVOKE UPDATE ON bell.site_quota_v1 FROM yovision_t011_sense`)
 	}()
 	value, err := OpenPostgres(ctx, os.Getenv(postgresTestDSNEnv))
 	if value != nil {
@@ -596,12 +597,12 @@ func TestPostgresOpenRejectsAreaSourcePrivilege(t *testing.T) {
 	_, admin := openPostgresTestStore(t)
 	ctx := context.Background()
 	if _, err := admin.ExecContext(ctx,
-		`GRANT SELECT ON bell.areas TO yovision_t010_sense`); err != nil {
+		`GRANT SELECT ON bell.areas TO yovision_t011_sense`); err != nil {
 		t.Fatal(err)
 	}
 	defer func() {
 		_, _ = admin.ExecContext(context.Background(),
-			`REVOKE SELECT ON bell.areas FROM yovision_t010_sense`)
+			`REVOKE SELECT ON bell.areas FROM yovision_t011_sense`)
 	}()
 	value, err := OpenPostgres(ctx, os.Getenv(postgresTestDSNEnv))
 	if value != nil {
@@ -613,6 +614,361 @@ func TestPostgresOpenRejectsAreaSourcePrivilege(t *testing.T) {
 	}
 }

+func TestPostgresOpenRejectsPublicControlStatePrivilege(t *testing.T) {
+	_, admin := openPostgresTestStore(t)
+	ctx := context.Background()
+	if _, err := admin.ExecContext(ctx, `GRANT SELECT ON sense.batch_operations TO PUBLIC`); err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		_, _ = admin.ExecContext(context.Background(), `REVOKE SELECT ON sense.batch_operations FROM PUBLIC`)
+	}()
+	value, err := OpenPostgres(ctx, os.Getenv(postgresTestDSNEnv))
+	if value != nil {
+		_ = value.Close()
+		t.Fatal("PUBLIC Control API table privilege was accepted")
+	}
+	if err == nil || !strings.Contains(err.Error(), "Control API state privilege boundary") {
+		t.Fatalf("expected Control API privilege-boundary error, got %v", err)
+	}
+}
+
+func TestPostgresControlCreateIdempotencyAndRedactedSnapshot(t *testing.T) {
+	postgres, admin := openPostgresTestStore(t)
+	insertBellSite(t, admin, "tenant", "site", 2)
+	value := videoDevice(1, "tenant", "site")
+	value.ID = "dev_control_create_1"
+	value.DesiredState = device.DesiredDisabled
+	hash := sha256.Sum256([]byte("canonical-create"))
+	ctx := WithAuditContext(context.Background(), AuditContext{
+		ActorType: AuditActorService, ActorID: "bell-control", TraceID: "trace-control-create",
+	})
+	request := ControlCreateRequest{
+		Scope: IdempotencyScope{
+			PrincipalID: "bell-control", TenantID: "tenant", SiteID: "site",
+			Operation: "createDevice", Key: "create-control-0001",
+			RequestHash: hash, TraceID: "trace-control-create",
+		},
+		Device: value,
+	}
+	first, err := postgres.CreateControlDevice(ctx, request)
+	if err != nil {
+		t.Fatal(err)
+	}
+	request.Device.ID = "dev_control_create_retry"
+	replayed, err := postgres.CreateControlDevice(ctx, request)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !replayed.Replay || replayed.Device.ID != first.Device.ID || replayed.TraceID != first.TraceID {
+		t.Fatalf("create replay drifted: first=%+v replay=%+v", first, replayed)
+	}
+	conflictHash := sha256.Sum256([]byte("different-create"))
+	request.Scope.RequestHash = conflictHash
+	if _, err := postgres.CreateControlDevice(ctx, request); !errors.Is(err, ErrIdempotencyConflict) {
+		t.Fatalf("same key with different body was not rejected: %v", err)
+	}
+	var devices, receipts int
+	var responseBody string
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.devices`).Scan(&devices); err != nil {
+		t.Fatal(err)
+	}
+	if err := admin.QueryRow(`SELECT count(*), min(response_body::text)
+		FROM sense.control_idempotency_receipts`).Scan(&receipts, &responseBody); err != nil {
+		t.Fatal(err)
+	}
+	if devices != 1 || receipts != 1 {
+		t.Fatalf("idempotent create counts drifted: devices=%d receipts=%d", devices, receipts)
+	}
+	for _, forbidden := range []string{value.EndpointRef, value.CredentialRef, "profile_token", "path_name"} {
+		if strings.Contains(responseBody, forbidden) {
+			t.Fatalf("idempotency response snapshot leaked %q", forbidden)
+		}
+	}
+}
+
+func TestPostgresControlListIsStableFilteredAndTenantScoped(t *testing.T) {
+	postgres, admin := openPostgresTestStore(t)
+	insertBellSite(t, admin, "tenant", "site", 4)
+	insertBellSite(t, admin, "other", "site", 4)
+	for index := 1; index <= 3; index++ {
+		value := videoDevice(index, "tenant", "site")
+		value.ID = fmt.Sprintf("dev_list_%d", index)
+		value.DesiredState = device.DesiredDisabled
+		if index == 3 {
+			value.DesiredState = device.DesiredEnabled
+		}
+		if err := postgres.CreateDevice(context.Background(), value); err != nil {
+			t.Fatal(err)
+		}
+	}
+	other := videoDevice(9, "other", "site")
+	other.ID, other.DesiredState = "dev_list_other", device.DesiredDisabled
+	if err := postgres.CreateDevice(context.Background(), other); err != nil {
+		t.Fatal(err)
+	}
+	first, err := postgres.ListControlDevices(context.Background(), "tenant", "site", ControlListFilter{Limit: 1})
+	if err != nil || len(first.Items) != 1 || !first.HasMore || first.Quota.UsedVideoChannels != 1 {
+		t.Fatalf("unexpected first page: %+v %v", first, err)
+	}
+	second, err := postgres.ListControlDevices(context.Background(), "tenant", "site", ControlListFilter{
+		Limit: 1, AfterCreated: &first.Items[0].CreatedAt, AfterDeviceID: first.Items[0].ID,
+	})
+	if err != nil || len(second.Items) != 1 || second.Items[0].ID == first.Items[0].ID {
+		t.Fatalf("stable cursor position failed: %+v %v", second, err)
+	}
+	desired := device.DesiredEnabled
+	filtered, err := postgres.ListControlDevices(context.Background(), "tenant", "site", ControlListFilter{
+		Limit: 100, DesiredState: &desired,
+	})
+	if err != nil || len(filtered.Items) != 1 || filtered.Items[0].ID != "dev_list_3" {
+		t.Fatalf("desired-state filter or tenant scope failed: %+v %v", filtered, err)
+	}
+	if _, err := postgres.ListControlDevices(context.Background(), "tenant", "missing", ControlListFilter{Limit: 50}); !errors.Is(err, ErrNotFound) {
+		t.Fatalf("missing Site did not return not found: %v", err)
+	}
+}
+
+func TestPostgresControlPatchUsesETagAndAuditV2(t *testing.T) {
+	postgres, admin := openPostgresTestStore(t)
+	insertBellSite(t, admin, "tenant", "site", 2)
+	value := videoDevice(1, "tenant", "site")
+	value.ID = "dev_control_patch_1"
+	value.DesiredState = device.DesiredDisabled
+	if err := postgres.CreateDevice(context.Background(), value); err != nil {
+		t.Fatal(err)
+	}
+	current, err := postgres.GetControlDevice(context.Background(), "tenant", "site", value.ID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	etag := DeviceETag(current.ID, current.ResourceVersion)
+	newName, newProfile := "Updated camera", "profile-main"
+	ctx := WithAuditContext(context.Background(), AuditContext{
+		ActorType: AuditActorUser, ActorID: "operator-1", TraceID: "trace-control-patch",
+	})
+	updated, err := postgres.PatchControlDevice(ctx, "tenant", "site", value.ID, etag, ControlPatch{
+		Name: &newName, ProfileToken: &newProfile,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if updated.Device.ResourceVersion != current.ResourceVersion+1 ||
+		updated.Device.Generation != current.Generation+1 || updated.ETag == etag {
+		t.Fatalf("patch did not advance versions: before=%+v after=%+v", current, updated)
+	}
+	if _, err := postgres.PatchControlDevice(ctx, "tenant", "site", value.ID, etag, ControlPatch{Name: &newName}); !errors.Is(err, ErrETagMismatch) {
+		t.Fatalf("stale ETag was accepted: %v", err)
+	}
+	var profileToken, eventType, payload string
+	if err := admin.QueryRow(`SELECT profile_token FROM sense.devices WHERE id = $1`, value.ID).Scan(&profileToken); err != nil {
+		t.Fatal(err)
+	}
+	if err := admin.QueryRow(`SELECT event_type, payload::text
+		FROM sense.device_operation_outbox WHERE event_type = 'device.configuration.accepted'`).
+		Scan(&eventType, &payload); err != nil {
+		t.Fatal(err)
+	}
+	if profileToken != newProfile || eventType != "device.configuration.accepted" ||
+		strings.Contains(payload, newProfile) || !strings.Contains(payload, "profile_token") {
+		t.Fatalf("configuration persistence/audit mismatch: profile=%q event=%q payload=%s", profileToken, eventType, payload)
+	}
+}
+
+func TestPostgresControlBatchIsPerItemDurableAndReplayable(t *testing.T) {
+	postgres, admin := openPostgresTestStore(t)
+	insertBellSite(t, admin, "tenant", "site", 4)
+	first := videoDevice(1, "tenant", "site")
+	first.ID, first.DesiredState = "dev_batch_1", device.DesiredDisabled
+	second := videoDevice(2, "tenant", "site")
+	second.ID, second.DesiredState = "dev_batch_2", device.DesiredDisabled
+	for _, value := range []device.Device{first, second} {
+		if err := postgres.CreateDevice(context.Background(), value); err != nil {
+			t.Fatal(err)
+		}
+	}
+	hash := sha256.Sum256([]byte("canonical-batch"))
+	ctx := WithAuditContext(context.Background(), AuditContext{
+		ActorType: AuditActorUser, ActorID: "operator-1", Reason: "approved", TraceID: "trace-batch",
+	})
+	request := ControlBatchRequest{
+		Scope: IdempotencyScope{
+			PrincipalID: "operator-1", TenantID: "tenant", SiteID: "site",
+			Operation: "batchSetDeviceDesiredState", Key: "batch-control-0001",
+			RequestHash: hash, TraceID: "trace-batch",
+		},
+		Reason: "approved",
+		Items: []ControlBatchItem{
+			{DeviceID: first.ID, ETag: DeviceETag(first.ID, 1), DesiredState: device.DesiredEnabled},
+			{DeviceID: first.ID, ETag: DeviceETag(first.ID, 1), DesiredState: device.DesiredEnabled},
+			{DeviceID: second.ID, ETag: DeviceETag(second.ID, 1), DesiredState: device.DesiredEnabled},
+		},
+	}
+	operation, err := postgres.BatchSetControlDesiredState(ctx, request)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if operation.Status != "partially_succeeded" || len(operation.Results) != 3 ||
+		operation.Results[0].Status != "rejected" || operation.Results[1].Status != "rejected" ||
+		operation.Results[2].Status != "succeeded" {
+		t.Fatalf("unexpected batch result: %+v", operation)
+	}
+	replayed, err := postgres.BatchSetControlDesiredState(ctx, request)
+	if err != nil || !replayed.Replay || replayed.ID != operation.ID {
+		t.Fatalf("batch replay drifted: %+v %v", replayed, err)
+	}
+	read, err := postgres.GetControlOperation(context.Background(), "tenant", operation.ID)
+	if err != nil || read.SiteID != "site" || len(read.Results) != 3 {
+		t.Fatalf("stored operation could not be read: %+v %v", read, err)
+	}
+	if _, err := postgres.GetControlOperation(context.Background(), "other-tenant", operation.ID); !errors.Is(err, ErrNotFound) {
+		t.Fatalf("cross-tenant operation was visible: %v", err)
+	}
+	var enabled, operations, receipts int
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.devices WHERE desired_state = 'enabled'`).Scan(&enabled); err != nil {
+		t.Fatal(err)
+	}
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.batch_operations`).Scan(&operations); err != nil {
+		t.Fatal(err)
+	}
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.control_idempotency_receipts`).Scan(&receipts); err != nil {
+		t.Fatal(err)
+	}
+	if enabled != 1 || operations != 1 || receipts != 1 {
+		t.Fatalf("batch durability counts drifted: enabled=%d operations=%d receipts=%d", enabled, operations, receipts)
+	}
+}
+
+func TestPostgresConcurrentControlCreateExecutesOnce(t *testing.T) {
+	postgres, admin := openPostgresTestStore(t)
+	insertBellSite(t, admin, "tenant", "site", 2)
+	hash := sha256.Sum256([]byte("concurrent-create"))
+	start := make(chan struct{})
+	results := make(chan ControlCreateResult, 2)
+	errorsFound := make(chan error, 2)
+	var wait sync.WaitGroup
+	for index := 1; index <= 2; index++ {
+		wait.Add(1)
+		go func(index int) {
+			defer wait.Done()
+			value := videoDevice(index, "tenant", "site")
+			value.ID = fmt.Sprintf("dev_concurrent_%d", index)
+			value.SerialNumber = "same-semantic-serial"
+			value.DesiredState = device.DesiredDisabled
+			request := ControlCreateRequest{
+				Scope: IdempotencyScope{
+					PrincipalID: "service", TenantID: "tenant", SiteID: "site",
+					Operation: "createDevice", Key: "concurrent-create-0001",
+					RequestHash: hash, TraceID: fmt.Sprintf("trace-concurrent-%d", index),
+				}, Device: value,
+			}
+			<-start
+			result, err := postgres.CreateControlDevice(context.Background(), request)
+			if err != nil {
+				errorsFound <- err
+				return
+			}
+			results <- result
+		}(index)
+	}
+	close(start)
+	wait.Wait()
+	close(results)
+	close(errorsFound)
+	for err := range errorsFound {
+		t.Fatalf("concurrent idempotent create failed: %v", err)
+	}
+	ids := make(map[string]struct{})
+	for result := range results {
+		ids[result.Device.ID] = struct{}{}
+	}
+	if len(ids) != 1 {
+		t.Fatalf("concurrent create returned multiple resources: %+v", ids)
+	}
+	var devices, audits, receipts int
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.devices`).Scan(&devices); err != nil {
+		t.Fatal(err)
+	}
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.device_operation_outbox`).Scan(&audits); err != nil {
+		t.Fatal(err)
+	}
+	if err := admin.QueryRow(`SELECT count(*) FROM sense.control_idempotency_receipts`).Scan(&receipts); err != nil {
+		t.Fatal(err)
+	}
+	if devices != 1 || audits != 1 || receipts != 1 {
+		t.Fatalf("concurrent create executed more than once: devices=%d audits=%d receipts=%d", devices, audits, receipts)
+	}
+}
+
+func TestPostgresConcurrentControlBatchesUseStableDeviceLockOrder(t *testing.T) {
+	postgres, admin := openPostgresTestStore(t)
+	insertBellSite(t, admin, "tenant", "site", 4)
+	ids := []string{"dev_lock_a", "dev_lock_b"}
+	for index, id := range ids {
+		value := videoDevice(index+1, "tenant", "site")
+		value.ID, value.DesiredState = id, device.DesiredDisabled
+		if err := postgres.CreateDevice(context.Background(), value); err != nil {
+			t.Fatal(err)
+		}
+	}
+	start := make(chan struct{})
+	operations := make(chan ControlBatchOperation, 2)
+	errorsFound := make(chan error, 2)
+	var wait sync.WaitGroup
+	for index := 0; index < 2; index++ {
+		wait.Add(1)
+		go func(index int) {
+			defer wait.Done()
+			order := ids
+			if index == 1 {
+				order = []string{ids[1], ids[0]}
+			}
+			hash := sha256.Sum256([]byte(fmt.Sprintf("batch-order-%d", index)))
+			request := ControlBatchRequest{
+				Scope: IdempotencyScope{
+					PrincipalID: "operator", TenantID: "tenant", SiteID: "site",
+					Operation: "batchSetDeviceDesiredState",
+					Key:       fmt.Sprintf("batch-lock-order-%04d", index), RequestHash: hash,
+					TraceID: fmt.Sprintf("trace-lock-order-%d", index),
+				},
+				Reason: "concurrency test",
+				Items: []ControlBatchItem{
+					{DeviceID: order[0], ETag: DeviceETag(order[0], 1), DesiredState: device.DesiredEnabled},
+					{DeviceID: order[1], ETag: DeviceETag(order[1], 1), DesiredState: device.DesiredEnabled},
+				},
+			}
+			<-start
+			operation, err := postgres.BatchSetControlDesiredState(context.Background(), request)
+			if err != nil {
+				errorsFound <- err
+				return
+			}
+			operations <- operation
+		}(index)
+	}
+	close(start)
+	wait.Wait()
+	close(operations)
+	close(errorsFound)
+	for err := range errorsFound {
+		t.Fatalf("opposite-order batch failed or deadlocked: %v", err)
+	}
+	var succeeded, failed int
+	for operation := range operations {
+		switch operation.Status {
+		case "succeeded":
+			succeeded++
+		case "failed":
+			failed++
+		default:
+			t.Fatalf("unexpected concurrent batch status: %+v", operation)
+		}
+	}
+	if succeeded != 1 || failed != 1 {
+		t.Fatalf("expected one winner and one stale loser, got succeeded=%d failed=%d", succeeded, failed)
+	}
+}
+
 func openPostgresTestStore(t *testing.T) (*Postgres, *sql.DB) {
 	t.Helper()
 	dsn := os.Getenv(postgresTestDSNEnv)
@@ -629,6 +985,9 @@ func openPostgresTestStore(t *testing.T) (*Postgres, *sql.DB) {
 		t.Fatal("connect PostgreSQL test administrator")
 	}
 	if _, err := admin.ExecContext(context.Background(), `TRUNCATE
+		sense.control_idempotency_receipts,
+		sense.batch_operation_items,
+		sense.batch_operations,
 		sense.device_operation_outbox,
 		sense.device_capabilities,
 		sense.reconcile_state,