Database governance built for humans and agents — controlling changes and access across every major database.
Go cicddata-governancedata-maskingdata-securitydatabase-accessdatabase-governancedevopsdevsecopsflywaygitopsliquibasemongodbmysqloraclepostgresqlschema-migrationssecuritysql-clientsqlservertidb
refactor(store): remove lifecycle writer coordination (#21304)
p0nyef5dc39bce3f
diff --git a/backend/api/v1/saved_query_service.go b/backend/api/v1/saved_query_service.goindex 4d8d2e3d21..7011ec780c 100644--- a/backend/api/v1/saved_query_service.go+++ b/backend/api/v1/saved_query_service.go@@ -36,8 +36,8 @@ func NewSavedQueryService(store *store.Store, iamManager *iam.Manager) *SavedQue}// CreateSavedQuery creates a new saved query. The bb.savedQueries.create-// permission is enforced at the interceptor; the store's insert transaction-// fences against a concurrent project purge (active project required).+// permission is enforced at the interceptor; the store also requires the+// project to be active when it validates the insert.func (s *SavedQueryService) CreateSavedQuery(ctx context.Context,req *connect.Request[v1pb.CreateSavedQueryRequest],@@ -57,8 +57,8 @@ func (s *SavedQueryService) CreateSavedQuery(}// Resource resolution returns archived projects (GetProject reports a- // project regardless of its deleted state), so this is what rejects a- // create into one. The store's fence covers the window after it.+ // project regardless of its deleted state), so this rejects a create into+ // one before the store repeats the point-in-time validation.project, err := s.store.GetProject(ctx, &store.FindProjectMessage{Workspace: common.GetWorkspaceIDFromContext(ctx),ResourceID: &projectResourceID,@@ -92,7 +92,6 @@ func (s *SavedQueryService) CreateSavedQuery(savedQuery, err := s.store.CreateSavedQuery(ctx, convertToStoreSavedQueryMessage(projectResourceID, database, user.Email, folder, request.SavedQuery))if err != nil {if common.ErrorCode(err) == common.NotFound {- // The fence lost to an archive or purge racing this create.return nil, connect.NewError(connect.CodeNotFound, errors.Wrapf(err, "project %q not found", projectResourceID))}return nil, connect.NewError(connect.CodeInternal, errors.Errorf("failed to create saved query: %v", err))diff --git a/backend/component/review/approval.go b/backend/component/review/approval.goindex 65bf511586..a6a3a51fed 100644--- a/backend/component/review/approval.go+++ b/backend/component/review/approval.go@@ -152,9 +152,7 @@ func (a *ApprovalEvaluator) ApplyApprovalTemplate(ctx context.Context, input Appreturn nil, workflowWrap(ErrorInternal, err, "failed to acquire Plan review lock for approval finding")}}- // The existing Issue row is the project-lifecycle fence for this update:- // project purge deletes Issues before the Project. Approval may finish while- // a soft-deleted Project still exists, but cannot write after purge passes it.+ // Lock the Issue before comparing and updating its approval payload.lockedIssue, err := lockIssue(ctx, tx, input.ProjectID, input.IssueUID)if err != nil {return nil, errdiff --git a/backend/component/review/approval_lifecycle_test.go b/backend/component/review/approval_lifecycle_test.godeleted file mode 100644index 48ceee381a..0000000000--- a/backend/component/review/approval_lifecycle_test.go+++ /dev/null@@ -1,238 +0,0 @@-package review--import (- "context"- "database/sql"- "errors"- "strings"- "testing"- "time"-- "github.com/stretchr/testify/require"-- storepb "github.com/bytebase/bytebase/backend/generated-go/store"- "github.com/bytebase/bytebase/backend/store"-)--const approvalLifecycleAdvisoryLockID = 990138--type approvalLifecycleApplyOutcome struct {- result *ApplyApprovalTemplateResult- err error-}--func TestApplyApprovalTemplateAndProjectDeletionPurgeFirst(t *testing.T) {- ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)- defer cancel()- stores := setupWorkflowStore(ctx, t)- plan, issue := newApprovalLifecycleFixture(ctx, t, stores)- softDeleteApprovalLifecycleProject(ctx, t, stores)-- db := stores.GetDB()- lockConn, err := db.Conn(ctx)- require.NoError(t, err)- defer lockConn.Close()- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", approvalLifecycleAdvisoryLockID)- require.NoError(t, err)- lockReleased := false- defer func() {- if !lockReleased {- _, _ = lockConn.ExecContext(context.Background(), "SELECT pg_advisory_unlock($1)", approvalLifecycleAdvisoryLockID)- }- }()- _, err = db.ExecContext(ctx, `- CREATE FUNCTION block_approval_lifecycle_issue_delete() RETURNS trigger AS $$- BEGIN- PERFORM pg_advisory_xact_lock(990138);- RETURN OLD;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER block_approval_lifecycle_issue_delete- AFTER DELETE ON issue- FOR EACH ROW EXECUTE FUNCTION block_approval_lifecycle_issue_delete();- `)- require.NoError(t, err)-- deleteResult := make(chan error, 1)- go func() {- deleteResult <- stores.DeleteProject(ctx, "default", "project-a")- }()- deleteBackendPID := waitForAdvisoryLockWaiter(ctx, t, db, approvalLifecycleAdvisoryLockID)-- applyReady := make(chan struct{})- applyResult := startApprovalLifecycleApply(ctx, stores, issue, applyReady)- <-applyReady- waitForBackendBlockedBy(ctx, t, db, deleteBackendPID,- "approval should wait for the Issue row being purged")-- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_unlock($1)", approvalLifecycleAdvisoryLockID)- require.NoError(t, err)- lockReleased = true- require.NoError(t, <-deleteResult)- assertApprovalLifecycleRejected(t, <-applyResult)- assertApprovalLifecyclePurged(ctx, t, stores, plan, issue)-}--func TestApplyApprovalTemplateAndProjectDeletionApprovalFirst(t *testing.T) {- ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)- defer cancel()- stores := setupWorkflowStore(ctx, t)- plan, issue := newApprovalLifecycleFixture(ctx, t, stores)- softDeleteApprovalLifecycleProject(ctx, t, stores)-- planCheckRunLock, planCheckRunLockPID := lockApprovalLifecyclePlanCheckRun(ctx, t, stores.GetDB(), plan)- lockReleased := false- defer func() {- if !lockReleased {- _ = planCheckRunLock.Rollback()- }- }()-- applyReady := make(chan struct{})- applyResult := startApprovalLifecycleApply(ctx, stores, issue, applyReady)- <-applyReady- applyBackendPID := waitForBackendBlockedBy(ctx, t, stores.GetDB(), planCheckRunLockPID,- "approval should lock the Issue, then wait for the Plan Check Run row")-- deleteResult := make(chan error, 1)- go func() {- deleteResult <- stores.DeleteProject(ctx, "default", "project-a")- }()- waitForBackendBlockedBy(ctx, t, stores.GetDB(), applyBackendPID,- "project deletion should wait for approval's Issue row lock")-- require.NoError(t, planCheckRunLock.Commit())- lockReleased = true- apply := <-applyResult- require.NoError(t, apply.err)- require.NotNil(t, apply.result)- require.True(t, apply.result.Applied)- require.NoError(t, <-deleteResult)- assertApprovalLifecyclePurged(ctx, t, stores, plan, issue)-}--func softDeleteApprovalLifecycleProject(ctx context.Context, t *testing.T, stores *store.Store) {- t.Helper()- deleted := true- require.NoError(t, stores.UpdateProjects(ctx, &store.UpdateProjectMessage{- Workspace: "default",- ResourceID: "project-a",- Delete: &deleted,- }))-}--func newApprovalLifecycleFixture(ctx context.Context, t *testing.T, stores *store.Store) (*store.PlanMessage, *store.IssueMessage) {- t.Helper()- plan, err := stores.CreatePlan(ctx, &store.PlanMessage{- ProjectID: "project-a",- Name: "approval lifecycle",- Config: &storepb.PlanConfig{ApprovalInputVersion: 2},- }, "[email protected]")- require.NoError(t, err)- issue, err := stores.CreateIssue(ctx, &store.IssueMessage{- ProjectID: "project-a",- CreatorEmail: "[email protected]",- Title: "approval lifecycle",- Type: storepb.Issue_DATABASE_CHANGE,- PlanUID: &plan.UID,- Payload: &storepb.Issue{Approval: &storepb.IssuePayloadApproval{- ApprovalInputVersion: 2,- }},- })- require.NoError(t, err)- created, err := stores.CreatePlanCheckRun(ctx, &store.PlanCheckRunMessage{- ProjectID: "project-a",- PlanUID: plan.UID,- Result: &storepb.PlanCheckRunResult{ApprovalInputVersion: 2},- })- require.NoError(t, err)- require.True(t, created)- planCheckRun, err := stores.GetPlanCheckRun(ctx, "project-a", plan.UID)- require.NoError(t, err)- require.NoError(t, stores.UpdatePlanCheckRun(ctx, "project-a", store.PlanCheckRunStatusDone, &storepb.PlanCheckRunResult{- ApprovalInputVersion: 2,- }, planCheckRun.UID))- return plan, issue-}--func startApprovalLifecycleApply(ctx context.Context, stores *store.Store, issue *store.IssueMessage, ready chan<- struct{}) <-chan approvalLifecycleApplyOutcome {- evaluator := &ApprovalEvaluator{workflow: NewWorkflow(stores)}- evaluator.evaluateApproval = func(_ context.Context, issue *store.IssueMessage, _ *store.ProjectMessage, _ *storepb.WorkspaceApprovalSetting) error {- issue.Payload.Approval = &storepb.IssuePayloadApproval{- ApprovalFindingDone: true,- ApprovalInputVersion: 2,- }- return nil- }- evaluator.beforeCommit = func() { close(ready) }- result := make(chan approvalLifecycleApplyOutcome, 1)- go func() {- applied, err := evaluator.ApplyApprovalTemplate(ctx, ApplyApprovalTemplateInput{- Workspace: "default",- ProjectID: issue.ProjectID,- IssueUID: issue.UID,- })- result <- approvalLifecycleApplyOutcome{result: applied, err: err}- }()- return result-}--func lockApprovalLifecyclePlanCheckRun(ctx context.Context, t *testing.T, db *sql.DB, plan *store.PlanMessage) (*sql.Tx, int) {- t.Helper()- tx, err := db.BeginTx(ctx, nil)- require.NoError(t, err)- var backendPID int- var planCheckRunUID int64- require.NoError(t, tx.QueryRowContext(ctx, `- SELECT pg_backend_pid(), id- FROM plan_check_run- WHERE project = $1- AND plan_id = $2- FOR UPDATE- `, plan.ProjectID, plan.UID).Scan(&backendPID, &planCheckRunUID))- require.NotZero(t, planCheckRunUID)- return tx, backendPID-}--func assertApprovalLifecycleRejected(t *testing.T, outcome approvalLifecycleApplyOutcome) {- t.Helper()- require.Nil(t, outcome.result)- var workflowErr *Error- require.Error(t, outcome.err)- require.True(t, errors.As(outcome.err, &workflowErr))- require.Equal(t, ErrorNotFound, workflowErr.Code)- require.NotContains(t, strings.ToLower(outcome.err.Error()), "foreign key")- require.NotContains(t, outcome.err.Error(), "23503")- require.NotContains(t, outcome.err.Error(), "40P01")-}--func assertApprovalLifecyclePurged(ctx context.Context, t *testing.T, stores *store.Store, plan *store.PlanMessage, issue *store.IssueMessage) {- t.Helper()- projectID := "project-a"- project, err := stores.GetProject(ctx, &store.FindProjectMessage{- Workspace: "default",- ResourceID: &projectID,- })- require.NoError(t, err)- require.Nil(t, project)-- gotIssue, err := stores.GetIssue(ctx, &store.FindIssueMessage{- Workspace: "default",- ProjectIDs: []string{projectID},- UID: &issue.UID,- })- require.NoError(t, err)- require.Nil(t, gotIssue)-- gotPlan, err := stores.GetPlan(ctx, &store.FindPlanMessage{- Workspace: "default",- ProjectID: projectID,- UID: &plan.UID,- })- require.NoError(t, err)- require.Nil(t, gotPlan)-- planCheckRun, err := stores.GetPlanCheckRun(ctx, projectID, plan.UID)- require.NoError(t, err)- require.Nil(t, planCheckRun)-}diff --git a/backend/component/review/draft_creation_lock_order_test.go b/backend/component/review/draft_creation_lock_order_test.godeleted file mode 100644index d9c6caac4f..0000000000--- a/backend/component/review/draft_creation_lock_order_test.go+++ /dev/null@@ -1,216 +0,0 @@-package review--import (- "context"- "database/sql"- "errors"- "testing"- "time"-- "github.com/stretchr/testify/require"-- storepb "github.com/bytebase/bytebase/backend/generated-go/store"- "github.com/bytebase/bytebase/backend/store"-)--const draftIssueDeletionAdvisoryLockID = 990137--type draftIssueDeletionFixture struct {- ctx context.Context- stores *store.Store- planUID int64-}--type createDraftIssueOutcome struct {- result *CreateDraftIssueResult- err error-}--func TestCreateDraftIssueAndProjectDeletionPurgeFirst(t *testing.T) {- fixture := newDraftIssueDeletionFixture(t)- ctx, db, stores := fixture.ctx, fixture.stores.GetDB(), fixture.stores-- lockConn, err := db.Conn(ctx)- require.NoError(t, err)- defer lockConn.Close()- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", draftIssueDeletionAdvisoryLockID)- require.NoError(t, err)- lockReleased := false- defer func() {- if !lockReleased {- _, _ = lockConn.ExecContext(context.Background(), "SELECT pg_advisory_unlock($1)", draftIssueDeletionAdvisoryLockID)- }- }()- _, err = db.ExecContext(ctx, `- CREATE FUNCTION block_draft_issue_plan_delete() RETURNS trigger AS $$- BEGIN- PERFORM pg_advisory_xact_lock(990137);- RETURN OLD;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER block_draft_issue_plan_delete- AFTER DELETE ON plan- FOR EACH ROW EXECUTE FUNCTION block_draft_issue_plan_delete();- `)- require.NoError(t, err)-- deleteResult := make(chan error, 1)- go func() {- deleteResult <- stores.DeleteProject(ctx, "default", "project-a")- }()- deleteBackendPID := waitForAdvisoryLockWaiter(ctx, t, db, draftIssueDeletionAdvisoryLockID)-- createResult := make(chan createDraftIssueOutcome, 1)- go func() {- result, err := createDraftIssue(fixture)- createResult <- createDraftIssueOutcome{result: result, err: err}- }()- waitForBackendBlockedBy(ctx, t, db, deleteBackendPID, "draft Issue creation should wait for the Plan being purged")-- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_unlock($1)", draftIssueDeletionAdvisoryLockID)- require.NoError(t, err)- lockReleased = true- require.NoError(t, <-deleteResult)- assertDraftIssueCreationNotFound(t, <-createResult)- assertDraftIssueProjectPurged(t, fixture)-}--func TestCreateDraftIssueAndProjectDeletionCreationFirst(t *testing.T) {- fixture := newDraftIssueDeletionFixture(t)- ctx, db, stores := fixture.ctx, fixture.stores.GetDB(), fixture.stores-- lockConn, err := db.Conn(ctx)- require.NoError(t, err)- defer lockConn.Close()- lockTx, err := lockConn.BeginTx(ctx, nil)- require.NoError(t, err)- lockReleased := false- defer func() {- if !lockReleased {- _ = lockTx.Rollback()- }- }()- var lockBackendPID int- require.NoError(t, lockTx.QueryRowContext(ctx, "SELECT pg_backend_pid()").Scan(&lockBackendPID))- var projectID string- require.NoError(t, lockTx.QueryRowContext(ctx, `- SELECT resource_id- FROM project- WHERE workspace = 'default' AND resource_id = 'project-a'- FOR UPDATE- `).Scan(&projectID))- require.Equal(t, "project-a", projectID)-- createResult := make(chan createDraftIssueOutcome, 1)- go func() {- result, err := createDraftIssue(fixture)- createResult <- createDraftIssueOutcome{result: result, err: err}- }()- createBackendPID := waitForBackendBlockedBy(ctx, t, db, lockBackendPID, "draft Issue creation should lock the Plan, then wait for the project")-- deleteResult := make(chan error, 1)- go func() {- deleteResult <- stores.DeleteProject(ctx, "default", "project-a")- }()- waitForBackendBlockedBy(ctx, t, db, createBackendPID, "project deletion should pass Issue cleanup, then wait for the locked Plan")-- require.NoError(t, lockTx.Commit())- lockReleased = true- assertDraftIssueCreationNotFound(t, <-createResult)- require.NoError(t, <-deleteResult)- assertDraftIssueProjectPurged(t, fixture)-}--func newDraftIssueDeletionFixture(t *testing.T) *draftIssueDeletionFixture {- t.Helper()- ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)- t.Cleanup(cancel)- stores := setupWorkflowStore(ctx, t)- plan, err := stores.CreatePlan(ctx, &store.PlanMessage{- ProjectID: "project-a",- Name: "draft Plan",- Config: &storepb.PlanConfig{Specs: []*storepb.PlanConfig_Spec{{- Config: &storepb.PlanConfig_Spec_ChangeDatabaseConfig{- ChangeDatabaseConfig: &storepb.PlanConfig_ChangeDatabaseConfig{},- },- }}},- }, "[email protected]")- require.NoError(t, err)- deleted := true- require.NoError(t, stores.UpdateProjects(ctx, &store.UpdateProjectMessage{- Workspace: "default", ResourceID: "project-a", Delete: &deleted,- }))- return &draftIssueDeletionFixture{ctx: ctx, stores: stores, planUID: plan.UID}-}--func createDraftIssue(fixture *draftIssueDeletionFixture) (*CreateDraftIssueResult, error) {- return NewWorkflow(fixture.stores).CreateDraftIssue(fixture.ctx, CreateDraftIssueInput{- Workspace: "default",- Issue: &store.IssueMessage{- ProjectID: "project-a", CreatorEmail: "[email protected]", PlanUID: &fixture.planUID,- Type: storepb.Issue_DATABASE_CHANGE,- Payload: &storepb.Issue{- Draft: true, Approval: &storepb.IssuePayloadApproval{},- },- },- })-}--func waitForAdvisoryLockWaiter(ctx context.Context, t *testing.T, db *sql.DB, lockID int) int {- t.Helper()- var backendPID int- require.Eventually(t, func() bool {- return db.QueryRowContext(ctx, `- SELECT COALESCE((- SELECT pid- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- ORDER BY pid- LIMIT 1- ), 0)- `, lockID).Scan(&backendPID) == nil && backendPID != 0- }, 10*time.Second, 10*time.Millisecond, "project deletion should reach the blocked Plan delete")- return backendPID-}--func waitForBackendBlockedBy(ctx context.Context, t *testing.T, db *sql.DB, blockerPID int, message string) int {- t.Helper()- var backendPID int- require.Eventually(t, func() bool {- return db.QueryRowContext(ctx, `- SELECT COALESCE((- SELECT pid- FROM pg_stat_activity- WHERE $1 = ANY(pg_blocking_pids(pid))- ORDER BY pid- LIMIT 1- ), 0)- `, blockerPID).Scan(&backendPID) == nil && backendPID != 0- }, 10*time.Second, 10*time.Millisecond, message)- return backendPID-}--func assertDraftIssueCreationNotFound(t *testing.T, outcome createDraftIssueOutcome) {- t.Helper()- require.Nil(t, outcome.result)- var workflowErr *Error- require.True(t, errors.As(outcome.err, &workflowErr))- require.Equal(t, ErrorNotFound, workflowErr.Code)-}--func assertDraftIssueProjectPurged(t *testing.T, fixture *draftIssueDeletionFixture) {- t.Helper()- projectID := "project-a"- project, err := fixture.stores.GetProject(fixture.ctx, &store.FindProjectMessage{- Workspace: "default", ResourceID: &projectID,- })- require.NoError(t, err)- require.Nil(t, project)- issue, err := fixture.stores.GetIssue(fixture.ctx, &store.FindIssueMessage{- Workspace: "default", ProjectIDs: []string{projectID}, PlanUID: &fixture.planUID,- })- require.NoError(t, err)- require.Nil(t, issue)-}diff --git a/backend/store/AGENTS.md b/backend/store/AGENTS.mdindex 69da6d8a78..bbab7bce34 100644--- a/backend/store/AGENTS.md+++ b/backend/store/AGENTS.md@@ -156,43 +156,25 @@ PostgreSQL holds row locks until a transaction ends. Transactions that acquire t- `changelog -> sync_history -> db -> instance -> project`- `revision -> db -> instance -> project`- `db_schema -> db -> instance -> project`++ These workflow chains are writer-to-writer correctness invariants. The+ best-effort lifecycle policy below does not relax them; in particular,+ preserve `issue -> plan` and `plan_check_run -> plan` ordering.3. Identify project-scoped rows with every scope column plus either the remaining primary-key columns or every remaining column of a declared non-partial unique key. Verify alternate keys in `LATEST.sql`. Lock batches in full primary-key order; project-scoped `(project, id)` batches therefore use that order, not `id` alone.4. Treat locks acquired by `UPDATE`, `DELETE`, foreign-key checks, and `INSERT ... ON CONFLICT DO UPDATE` as part of the order. An upsert that can update an existing row is not a new-row-only insert.5. `nextProjectID` locks `project` and requires it to be active before allocating an ID. Call it after locking any existing descendants, and do not lock an existing descendant afterward. Creation is rejected when the project is missing or deleted.-Row ordering prevents wait-for cycles on existing rows. It cannot protect an-absent child row because there is no row to lock before a concurrent purge passes-that branch. The active-project check in `nextProjectID` covers this case only for-writers that call it; it is not a repository-wide purge fence because other-writers bypass `nextProjectID`.--Database creation, database-sync, batch database updates, task-run creation,-sheet creation, and Query History writers, together with direct instance-archive and direct project/instance purge, additionally take the matching-transaction-scoped purge fence before any row lock. This closes absent-descendant gaps; writers then-retain the normal child-to-parent row-lock order. Database sync may continue for-an archived project while its row exists, but never through a soft-deleted-instance. Direct instance archive and restore fail while any targeting task run-is pending, available, or running.--Every new or modified writer of purge-managed data must define its project-lifecycle policy: require an active project for new resources, or require only an-existing project when deleted-project continuation is intentional. Serialize and-validate that policy against project deletion before writing the managed data.--Transactions spanning project- or instance-owned sibling branches follow this canonical order:--```text-query_history -> policy -> saved_query_organizer -> saved_query--> issue_comment -> review_run -> issue -> plan_webhook_delivery -> plan_check_run--> task_run_log -> task_run -> task -> plan -> access_grant -> release--> db_group -> changelog -> sync_history -> revision -> db_schema -> db--> sheet_blob_ref -> project_webhook -> service_account -> workload_identity -> instance -> project-```+Row ordering prevents wait-for cycles on existing rows. Keep transactions short+and preserve this order whether locks are acquired explicitly or by `UPDATE` and+`DELETE` statements.-Update this list, `DeleteProject`, and `DeleteInstance` together. A transaction that needs another sibling branch must establish its position here before implementation. When one table is touched by multiple predicates, keep those mutations contiguous at that table's position. Keep transactions short and preserve this order whether locks are acquired explicitly or by `UPDATE` and `DELETE` statements.+Project and instance archive and physical purge are point-in-time lifecycle+operations. Writers validate the current state required for their own work, but+do not coordinate with archive or purge. A rare concurrent writer may fail with+a PostgreSQL conflict or commit based on state it observed before the lifecycle+transition; these are accepted best-effort outcomes and the user may retry.-Examples:+Examples of ordinary transaction ordering:- Pending Task Run creation: existing `task` rows ordered by `(project, id)`, then `project`, then new `task_run` rows.- Plan Check Run refresh: existing `plan_check_run`, then `plan`, then `project`, then the upsert.@@ -200,7 +182,5 @@ Examples:- Task skipping: existing `task` rows ordered by `(project, id)`; it does not lock `task_run` rows.When adding or changing a transaction that coordinates multiple rows or tables,-add deterministic real-PostgreSQL regression tests for both lock-acquisition-directions. Assert the terminal outcomes, including that neither direction ends-in a foreign-key failure; merely checking for the absence of SQLSTATE `40P01` is-insufficient.+add deterministic real-PostgreSQL regression tests for its ordinary contention+paths and assert the terminal outcomes.diff --git a/backend/store/advisory_lock.go b/backend/store/advisory_lock.goindex c8fda98a4a..a060664003 100644--- a/backend/store/advisory_lock.go+++ b/backend/store/advisory_lock.go@@ -26,12 +26,6 @@ const (// AdvisoryLockKeyPlanIssueRollout serializes Plan review changes, linked// Bytebase Issue creation, and Rollout creation for the same Plan.AdvisoryLockKeyPlanIssueRollout AdvisoryLockKey = 1005- // AdvisoryLockKeyProjectPurge serializes project-owned writes with project- // purge, including writes to an otherwise absent database descendant.- AdvisoryLockKeyProjectPurge AdvisoryLockKey = 1006- // AdvisoryLockKeyInstancePurge serializes instance lifecycle operations with- // descendant creation and purge when the descendant row may not exist yet.- AdvisoryLockKeyInstancePurge AdvisoryLockKey = 1007// AdvisoryLockKeyIamPolicy serializes compare-and-swap writes to one// resource's IAM policy, including the first write, when there is no policy// row to lock yet.@@ -45,14 +39,6 @@ func AcquirePlanIssueRolloutAdvisoryLock(ctx context.Context, tx *sql.Tx, projecreturn AcquireAdvisoryXactLockWithStringKey(ctx, tx, AdvisoryLockKeyPlanIssueRollout, key)}-func acquireProjectPurgeLock(ctx context.Context, tx *sql.Tx, projectID string) error {- return AcquireAdvisoryXactLockWithStringKey(ctx, tx, AdvisoryLockKeyProjectPurge, projectID)-}--func acquireInstancePurgeLock(ctx context.Context, tx *sql.Tx, instanceID string) error {- return AcquireAdvisoryXactLockWithStringKey(ctx, tx, AdvisoryLockKeyInstancePurge, instanceID)-}-// AdvisoryLock holds a dedicated connection for a session-level advisory lock.type AdvisoryLock struct {conn *sql.Conn@@ -70,12 +56,12 @@ func TryAdvisoryLock(ctx context.Context, db *sql.DB, key AdvisoryLockKey) (*Advvar acquired boolif err := conn.QueryRowContext(ctx, "SELECT pg_try_advisory_lock($1)", int64(key)).Scan(&acquired); err != nil {- conn.Close()+ _ = conn.Close()return nil, false, err}if !acquired {- conn.Close()+ _ = conn.Close()return nil, false, nil}@@ -143,7 +129,7 @@ func AcquireAdvisoryLock(ctx context.Context, db *sql.DB, key AdvisoryLockKey) (// pg_advisory_lock blocks until the lock is acquiredif _, err := conn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", int64(key)); err != nil {- conn.Close()+ _ = conn.Close()return nil, err}diff --git a/backend/store/batch_update_database_purge_test.go b/backend/store/batch_update_database_purge_test.godeleted file mode 100644index 91c8b353d2..0000000000--- a/backend/store/batch_update_database_purge_test.go+++ /dev/null@@ -1,250 +0,0 @@-package store_test--import (- "database/sql"- "strings"- "testing"-- "github.com/stretchr/testify/require"-- "github.com/bytebase/bytebase/backend/store"-)--func batchMoveDatabaseIntoProjectA(fixture *projectDeletionLockOrderFixture) error {- projectID := "project-a"- return fixture.store.BatchUpdateDatabases(fixture.ctx, []*store.DatabaseMessage{{- InstanceID: "instance-a",- DatabaseName: "db-a",- }}, &store.BatchUpdateDatabases{- Workspace: "default",- ProjectID: &projectID,- })-}--func requireProjectPurgeTerminalState(t *testing.T, fixture *projectDeletionLockOrderFixture) {- t.Helper()- var projectCount, orphanCount, databaseCount, sheetRefCount int- var project string- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM project WHERE resource_id = 'project-a'").Scan(&projectCount))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM db WHERE project = 'project-a'").Scan(&orphanCount))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM db WHERE instance = 'instance-a' AND name = 'db-a'").Scan(&databaseCount))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT project FROM db WHERE instance = 'instance-a' AND name = 'db-a'").Scan(&project))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM sheet_blob_ref WHERE project = 'project-a'").Scan(&sheetRefCount))- require.Zero(t, projectCount, "project purge must remove the archived project")- require.Zero(t, orphanCount, "no database row may reference the purged project")- require.Equal(t, 1, databaseCount, "the workspace-instance database must survive")- require.Equal(t, "default", project, "the workspace-instance database must be reassigned to the default project")- require.Zero(t, sheetRefCount, "no sheet ref may reference the purged project")-}--func TestBatchUpdateDatabasesAndDeleteProjectSerialize(t *testing.T) {- // The batch moves a workspace-instance database into the archived project- // while that project is being purged. Without purge fencing the terminal- // project deletion fails on db.project's foreign key. The archived project- // also owns a sheet ref, so the purge must clear sheet_blob_ref or fail on- // its foreign key.- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', NULL);- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'default');- INSERT INTO sheet_blob (sha256, content) VALUES (sha256('SELECT transfer;'::bytea), 'SELECT transfer;');- INSERT INTO sheet_blob_ref (project, sha256) VALUES ('project-a', sha256('SELECT transfer;'::bytea));- `- purgeProjectA := func(fixture *projectDeletionLockOrderFixture) error {- return fixture.store.DeleteProject(fixture.ctx, "default", "project-a")- }-- t.Run("purge first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9941- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER DELETE ON project FOR EACH ROW")-- purgeResult := make(chan error, 1)- go func() { purgeResult <- purgeProjectA(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- batchResult := make(chan error, 1)- go func() { batchResult <- batchMoveDatabaseIntoProjectA(fixture) }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, purgeResult)- err := <-batchResult- require.Error(t, err)- require.NotContains(t, strings.ToLower(err.Error()), "foreign key",- "the batch must reject the purged destination cleanly, not with an FK failure")- require.ErrorContains(t, err, "project-a not found",- "the batch must fail on the purged destination project")- requireProjectPurgeTerminalState(t, fixture)- })-- t.Run("batch first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9942- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- _, err := fixture.db.ExecContext(fixture.ctx, `- CREATE FUNCTION block_db_project_update_9942() RETURNS trigger AS $$- BEGIN- PERFORM pg_advisory_xact_lock(9942);- RETURN NEW;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER block_db_project_update_9942- BEFORE UPDATE OF project ON db FOR EACH ROW- WHEN (OLD.project IS DISTINCT FROM NEW.project)- EXECUTE FUNCTION block_db_project_update_9942();- `)- require.NoError(t, err)-- batchResult := make(chan error, 1)- go func() { batchResult <- batchMoveDatabaseIntoProjectA(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- purgeResult := make(chan error, 1)- go func() { purgeResult <- purgeProjectA(fixture) }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, batchResult)- requireMaintenanceResult(t, purgeResult)- requireProjectPurgeTerminalState(t, fixture)- })-}--func TestBatchUpdateDatabasesRejectsArchivedInstance(t *testing.T) {- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project, deleted)- VALUES ('instance-a', 'default', NULL, TRUE);- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'default');- `- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- err := batchMoveDatabaseIntoProjectA(fixture)- require.Error(t, err)- require.ErrorContains(t, err, "instance instance-a is archived")-- var projectID string- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT project FROM db WHERE instance = 'instance-a' AND name = 'db-a'").Scan(&projectID))- require.Equal(t, "default", projectID)-}--func TestBatchUpdateDatabasesClearsEnvironmentOnArchivedInstance(t *testing.T) {- environmentID, unset := "env-a", ""- fixture := newProjectDeletionLockOrderFixture(t, `- INSERT INTO instance (resource_id, workspace, project, deleted)- VALUES ('instance-a', 'default', NULL, TRUE);- INSERT INTO db (instance, name, project, environment)- VALUES ('instance-a', 'db-a', 'default', 'env-a');- `)-- err := fixture.store.BatchUpdateDatabases(fixture.ctx, nil, &store.BatchUpdateDatabases{- Workspace: "default",- FindByEnvironmentID: &environmentID,- EnvironmentID: &unset,- })- require.NoError(t, err, "environment cleanup must remain possible after an instance is archived")-- var environment sql.NullString- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT environment FROM db WHERE instance = 'instance-a' AND name = 'db-a'").Scan(&environment))- require.False(t, environment.Valid, "the archived instance must not retain a deleted environment reference")-}--func TestBatchUpdateDatabasesByEnvironment(t *testing.T) {- envA, unset := "env-a", ""- fixture := newProjectDeletionLockOrderFixture(t, `- INSERT INTO workspace (resource_id) VALUES ('other');- INSERT INTO instance (resource_id, workspace, project) VALUES- ('instance-a', 'default', NULL),- ('instance-b', 'other', NULL);- INSERT INTO db (instance, name, project, environment) VALUES- ('instance-a', 'db-a', 'default', 'env-a'),- ('instance-a', 'db-b', 'default', 'env-b'),- ('instance-b', 'db-c', 'default', 'env-a');- `)-- err := fixture.store.BatchUpdateDatabases(fixture.ctx, nil, &store.BatchUpdateDatabases{- Workspace: "default",- FindByEnvironmentID: &envA,- EnvironmentID: &unset,- })- require.NoError(t, err)-- requireDatabaseEnvironment := func(instanceID, databaseName string, want string) {- t.Helper()- var environment sql.NullString- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT environment FROM db WHERE instance = $1 AND name = $2", instanceID, databaseName).Scan(&environment))- if want == "" {- require.False(t, environment.Valid, "database %s/%s environment should be unset", instanceID, databaseName)- } else {- require.True(t, environment.Valid)- require.Equal(t, want, environment.String)- }- }- requireDatabaseEnvironment("instance-a", "db-a", "")- requireDatabaseEnvironment("instance-a", "db-b", "env-b")- requireDatabaseEnvironment("instance-b", "db-c", "env-a")-}--func TestBatchUpdateDatabasesByEnvironmentRevalidatesDynamicTargets(t *testing.T) {- environmentID, unset := "env-a", ""- fixture := newProjectDeletionLockOrderFixture(t, `- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', NULL);- INSERT INTO db (instance, name, project, environment) VALUES- ('instance-a', 'db-a', 'default', 'env-a'),- ('instance-a', 'db-b', 'default', 'env-b');- `)-- // Block the batch on its default-project purge fence. It reaches that fence- // only after the outside-transaction target read has completed.- conn, err := fixture.db.Conn(fixture.ctx)- require.NoError(t, err)- t.Cleanup(func() { require.NoError(t, conn.Close()) })- _, err = conn.ExecContext(fixture.ctx, "SELECT pg_advisory_lock($1, hashtext($2))", int64(store.AdvisoryLockKeyProjectPurge), "default")- require.NoError(t, err)- t.Cleanup(func() {- _, err := conn.ExecContext(fixture.ctx, "SELECT pg_advisory_unlock($1, hashtext($2))", int64(store.AdvisoryLockKeyProjectPurge), "default")- require.NoError(t, err)- })- var barrierPID int- require.NoError(t, conn.QueryRowContext(fixture.ctx, "SELECT pg_backend_pid()").Scan(&barrierPID))-- result := make(chan error, 1)- go func() {- result <- fixture.store.BatchUpdateDatabases(fixture.ctx, nil, &store.BatchUpdateDatabases{- Workspace: "default",- FindByEnvironmentID: &environmentID,- EnvironmentID: &unset,- })- }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, barrierPID)-- // db-b was not in the pre-read target set, but becomes a match before the- // fenced transaction scans. The batch must either include it or explicitly- // request a retry; it must not report success and silently leave it behind.- _, err = fixture.db.ExecContext(fixture.ctx,- "UPDATE db SET environment = 'env-a' WHERE instance = 'instance-a' AND name = 'db-b'")- require.NoError(t, err)- _, err = conn.ExecContext(fixture.ctx, "SELECT pg_advisory_unlock($1, hashtext($2))", int64(store.AdvisoryLockKeyProjectPurge), "default")- require.NoError(t, err)-- err = <-result- if err != nil {- require.NotContains(t, strings.ToLower(err.Error()), "foreign key")- require.Contains(t, strings.ToLower(err.Error()), "retry")- return- }-- var environment sql.NullString- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT environment FROM db WHERE instance = 'instance-a' AND name = 'db-b'").Scan(&environment))- require.False(t, environment.Valid, "a matching database added during the fence wait must not be skipped")-}diff --git a/backend/store/batch_update_database_test.go b/backend/store/batch_update_database_test.gonew file mode 100644index 0000000000..bd76f6594c--- /dev/null+++ b/backend/store/batch_update_database_test.go@@ -0,0 +1,97 @@+package store_test++import (+ "database/sql"+ "testing"++ "github.com/stretchr/testify/require"++ "github.com/bytebase/bytebase/backend/store"+)++func batchMoveDatabaseIntoProjectA(fixture *storePostgresFixture) error {+ projectID := "project-a"+ return fixture.store.BatchUpdateDatabases(fixture.ctx, []*store.DatabaseMessage{{+ InstanceID: "instance-a",+ DatabaseName: "db-a",+ }}, &store.BatchUpdateDatabases{+ Workspace: "default",+ ProjectID: &projectID,+ })+}++func TestBatchUpdateDatabasesRejectsArchivedInstance(t *testing.T) {+ const seedSQL = `+ INSERT INTO instance (resource_id, workspace, project, deleted)+ VALUES ('instance-a', 'default', NULL, TRUE);+ INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'default');+ `+ fixture := newStorePostgresFixture(t, seedSQL)+ err := batchMoveDatabaseIntoProjectA(fixture)+ require.Error(t, err)+ require.ErrorContains(t, err, "instance instance-a is archived")++ var projectID string+ require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,+ "SELECT project FROM db WHERE instance = 'instance-a' AND name = 'db-a'").Scan(&projectID))+ require.Equal(t, "default", projectID)+}++func TestBatchUpdateDatabasesClearsEnvironmentOnArchivedInstance(t *testing.T) {+ environmentID, unset := "env-a", ""+ fixture := newStorePostgresFixture(t, `+ INSERT INTO instance (resource_id, workspace, project, deleted)+ VALUES ('instance-a', 'default', NULL, TRUE);+ INSERT INTO db (instance, name, project, environment)+ VALUES ('instance-a', 'db-a', 'default', 'env-a');+ `)++ err := fixture.store.BatchUpdateDatabases(fixture.ctx, nil, &store.BatchUpdateDatabases{+ Workspace: "default",+ FindByEnvironmentID: &environmentID,+ EnvironmentID: &unset,+ })+ require.NoError(t, err, "environment cleanup must remain possible after an instance is archived")++ var environment sql.NullString+ require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,+ "SELECT environment FROM db WHERE instance = 'instance-a' AND name = 'db-a'").Scan(&environment))+ require.False(t, environment.Valid, "the archived instance must not retain a deleted environment reference")+}++func TestBatchUpdateDatabasesByEnvironment(t *testing.T) {+ envA, unset := "env-a", ""+ fixture := newStorePostgresFixture(t, `+ INSERT INTO workspace (resource_id) VALUES ('other');+ INSERT INTO instance (resource_id, workspace, project) VALUES+ ('instance-a', 'default', NULL),+ ('instance-b', 'other', NULL);+ INSERT INTO db (instance, name, project, environment) VALUES+ ('instance-a', 'db-a', 'default', 'env-a'),+ ('instance-a', 'db-b', 'default', 'env-b'),+ ('instance-b', 'db-c', 'default', 'env-a');+ `)++ err := fixture.store.BatchUpdateDatabases(fixture.ctx, nil, &store.BatchUpdateDatabases{+ Workspace: "default",+ FindByEnvironmentID: &envA,+ EnvironmentID: &unset,+ })+ require.NoError(t, err)++ requireDatabaseEnvironment := func(instanceID, databaseName string, want string) {+ t.Helper()+ var environment sql.NullString+ require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,+ "SELECT environment FROM db WHERE instance = $1 AND name = $2", instanceID, databaseName).Scan(&environment))+ if want == "" {+ require.False(t, environment.Valid, "database %s/%s environment should be unset", instanceID, databaseName)+ } else {+ require.True(t, environment.Valid)+ require.Equal(t, want, environment.String)+ }+ }+ requireDatabaseEnvironment("instance-a", "db-a", "")+ requireDatabaseEnvironment("instance-a", "db-b", "env-b")+ requireDatabaseEnvironment("instance-b", "db-c", "env-a")+}diff --git a/backend/store/changelog.go b/backend/store/changelog.goindex a18b23bd57..fb87aade44 100644--- a/backend/store/changelog.go+++ b/backend/store/changelog.go@@ -94,7 +94,7 @@ func (s *Store) CreateChangelog(ctx context.Context, create *ChangelogMessage) (}var resourceID string- err = s.withDatabasePurgeFence(ctx, create.InstanceID, create.DatabaseName, "", func(tx *sql.Tx) error {+ err = s.withDatabaseWrite(ctx, create.InstanceID, create.DatabaseName, func(tx *sql.Tx) error {if create.SyncHistory == nil {return nil}diff --git a/backend/store/database.go b/backend/store/database.goindex f61b5901af..c1567d449c 100644--- a/backend/store/database.go+++ b/backend/store/database.go@@ -280,7 +280,7 @@ func (s *Store) ListDatabases(ctx context.Context, find *FindDatabaseMessage) ([// CreateDatabaseDefault creates a database discovered by schema sync.func (s *Store) CreateDatabaseDefault(ctx context.Context, create *DatabaseMessage) (*DatabaseMessage, error) {- err := s.withDatabasePurgeFence(ctx, create.InstanceID, create.DatabaseName, create.ProjectID, nil, func(tx *sql.Tx, ownership *databaseOwnership) error {+ err := s.withDatabaseWrite(ctx, create.InstanceID, create.DatabaseName, nil, func(tx *sql.Tx, ownership *databaseOwnership) error {projectID, err := ownership.projectForDefaultCreate(create.ProjectID)if err != nil {return err@@ -312,7 +312,7 @@ func (s *Store) UpsertDatabase(ctx context.Context, create *DatabaseMessage) (*Dif create.EnvironmentID != nil && *create.EnvironmentID != "" {environment = create.EnvironmentID}- err = s.withDatabasePurgeFence(ctx, create.InstanceID, create.DatabaseName, create.ProjectID, nil, func(tx *sql.Tx, ownership *databaseOwnership) error {+ err = s.withDatabaseWrite(ctx, create.InstanceID, create.DatabaseName, nil, func(tx *sql.Tx, ownership *databaseOwnership) error {projectID, err := ownership.projectForUpsert(create.ProjectID)if err != nil {return err@@ -338,11 +338,7 @@ func (s *Store) UpsertDatabase(ctx context.Context, create *DatabaseMessage) (*D// UpdateDatabase updates a database.func (s *Store) UpdateDatabase(ctx context.Context, patch *UpdateDatabaseMessage) (*DatabaseMessage, error) {- requestedProjectID := ""- if patch.ProjectID != nil {- requestedProjectID = *patch.ProjectID- }- err := s.withDatabasePurgeFence(ctx, patch.InstanceID, patch.DatabaseName, requestedProjectID, nil, func(tx *sql.Tx, ownership *databaseOwnership) error {+ err := s.withDatabaseWrite(ctx, patch.InstanceID, patch.DatabaseName, nil, func(tx *sql.Tx, ownership *databaseOwnership) error {if !ownership.exists {return common.Errorf(common.NotFound, "database %s not found", common.FormatDatabase(patch.InstanceID, patch.DatabaseName))}@@ -396,12 +392,6 @@ func (s *Store) UpdateDatabase(ctx context.Context, patch *UpdateDatabaseMessage}// BatchUpdateDatabases updates databases in batch.-//-// The batch runs in one transaction after acquiring every affected source and-// destination project purge fence and instance purge fence in deterministic-// sorted order. Without the fences, a concurrent project purge could sweep the-// destination project's rows and make the terminal project deletion fail on-// the db.project foreign key.func (s *Store) BatchUpdateDatabases(ctx context.Context, databases []*DatabaseMessage, update *BatchUpdateDatabases) error {set := qb.Q()if update.ProjectID != nil {@@ -438,45 +428,12 @@ func (s *Store) BatchUpdateDatabases(ctx context.Context, databases []*DatabaseMwhere.And("db.instance IN (SELECT resource_id FROM instance WHERE workspace = ?)", update.Workspace)}- // Discover every affected database before acquiring purge fences so the- // source/destination project and instance fence sets are known up front.- // The transaction below re-reads these targets under the fences.- targets, err := s.listDatabaseBatchTargets(ctx, where)- if err != nil {- return err- }- fenceInstances := make([]string, 0, len(targets))- fenceProjects := make([]string, 0, len(targets))- for _, target := range targets {- fenceInstances = append(fenceInstances, target.instanceID)- fenceProjects = append(fenceProjects, target.projectID)- if target.instanceProject != "" {- fenceProjects = append(fenceProjects, target.instanceProject)- }- }- if update.ProjectID != nil {- fenceProjects = append(fenceProjects, *update.ProjectID)- }- slices.Sort(fenceInstances)- fenceInstances = slices.Compact(fenceInstances)- slices.Sort(fenceProjects)- fenceProjects = slices.Compact(fenceProjects)-var updated []databaseBatchTarget- err = s.withDatabaseBatchPurgeFence(ctx, fenceInstances, fenceProjects, func(tx *sql.Tx) error {- // Dynamic environment matching can gain rows between the pre-read and- // the purge fences. Lock the complete current match set so a newly- // matching row is either included or causes a retry below; never leave- // it silently unchanged.- locked, err := lockDatabaseBatchTargets(ctx, tx, where, targets, update.FindByEnvironmentID != nil)+ err := s.withDatabaseBatchWrite(ctx, func(tx *sql.Tx) error {+ locked, err := lockDatabaseBatchTargets(ctx, tx, where)if err != nil {return err}- for _, target := range locked {- if !slices.Contains(fenceInstances, target.instanceID) || !slices.Contains(fenceProjects, target.projectID) || (target.instanceProject != "" && !slices.Contains(fenceProjects, target.instanceProject)) {- return common.Errorf(common.Conflict, "batch database targets changed; retry")- }- }instances, err := lockDatabaseBatchInstances(ctx, tx, locked)if err != nil {return err@@ -488,42 +445,6 @@ func (s *Store) BatchUpdateDatabases(ctx context.Context, databases []*DatabaseM}}}- // Revalidate lifecycle under the transaction: every target must still- // be covered by the pre-read purge fences. A purge that committed- // before the fences were acquired shows up here as a retry instead of- // an FK failure.- for _, target := range locked {- if !slices.Contains(fenceProjects, target.projectID) {- return common.Errorf(common.Conflict, "database ownership changed to project %s for %s; retry", target.projectID, common.FormatDatabase(target.instanceID, target.databaseName))- }- if instanceProject := instances[target.instanceID].projectID; instanceProject != "" && !slices.Contains(fenceProjects, instanceProject) {- return common.Errorf(common.Conflict, "database ownership changed to project instance %s for %s; retry", instanceProject, common.FormatDatabase(target.instanceID, target.databaseName))- }- }- // Lock every affected project, including the destination, after the- // database rows and instances. A purged destination fails cleanly with- // NotFound instead of an FK violation.- projectIDs := make([]string, 0, len(locked)+1)- for _, target := range locked {- projectIDs = append(projectIDs, target.projectID)- if instanceProject := instances[target.instanceID].projectID; instanceProject != "" {- projectIDs = append(projectIDs, instanceProject)- }- }- if update.ProjectID != nil {- projectIDs = append(projectIDs, *update.ProjectID)- }- slices.Sort(projectIDs)- projectIDs = slices.Compact(projectIDs)- for _, projectID := range projectIDs {- var foundProjectID string- if err := tx.QueryRowContext(ctx, "SELECT resource_id FROM project WHERE resource_id = $1 FOR UPDATE", projectID).Scan(&foundProjectID); err != nil {- if errors.Is(err, sql.ErrNoRows) {- return common.Errorf(common.NotFound, "project %s not found", projectID)- }- return errors.Wrapf(err, "failed to lock project %s", projectID)- }- }// Atomic validation: any invalid target rejects the whole batch before// any row is updated.if update.ProjectID != nil {@@ -568,103 +489,32 @@ func (s *Store) BatchUpdateDatabases(ctx context.Context, databases []*DatabaseM}type databaseBatchTarget struct {- instanceID string- databaseName string- projectID string- instanceProject string-}--// listDatabaseBatchTargets finds the databases matched by the batch update-// where clause. It runs outside the purge fences; the result only determines-// which fences to acquire and which rows to lock inside the transaction.-func (s *Store) listDatabaseBatchTargets(ctx context.Context, where *qb.Query) ([]databaseBatchTarget, error) {- q := qb.Q().Space(`- SELECT db.instance, db.name, db.project, instance.project- FROM db- JOIN instance ON instance.resource_id = db.instance- WHERE ?- ORDER BY db.instance, db.name- `, where)- query, args, err := q.ToSQL()- if err != nil {- return nil, errors.Wrap(err, "failed to build batch database target query")- }- rows, err := s.GetDB().QueryContext(ctx, query, args...)- if err != nil {- return nil, errors.Wrap(err, "failed to find batch database targets")- }- defer rows.Close()- var targets []databaseBatchTarget- for rows.Next() {- var target databaseBatchTarget- var instanceProject sql.NullString- if err := rows.Scan(&target.instanceID, &target.databaseName, &target.projectID, &instanceProject); err != nil {- return nil, errors.Wrap(err, "failed to scan batch database target")- }- if instanceProject.Valid {- target.instanceProject = instanceProject.String- }- targets = append(targets, target)- }- if err := rows.Err(); err != nil {- return nil, errors.Wrap(err, "failed to read batch database targets")- }- return targets, nil+ instanceID string+ databaseName string+ projectID string}-// withDatabaseBatchPurgeFence serializes a batch database write with direct-// project and instance purge. Advisory fences are acquired in deterministic-// sorted order (all project fences, then all instance fences) before any row-// lock.-func (s *Store) withDatabaseBatchPurgeFence(- ctx context.Context,- instances []string,- projects []string,- write func(*sql.Tx) error,-) error {+func (s *Store) withDatabaseBatchWrite(ctx context.Context, write func(*sql.Tx) error) error {tx, err := s.GetDB().BeginTx(ctx, nil)if err != nil {- return errors.Wrap(err, "failed to begin batch database write transaction")+ return errors.Wrap(err, "failed to begin batch database transaction")}defer tx.Rollback()- for _, projectID := range projects {- if err := acquireProjectPurgeLock(ctx, tx, projectID); err != nil {- return errors.Wrapf(err, "failed to lock project purge fence for %s", projectID)- }- }- for _, instanceID := range instances {- if err := acquireInstancePurgeLock(ctx, tx, instanceID); err != nil {- return errors.Wrapf(err, "failed to lock instance purge fence for %s", instanceID)- }- }if err := write(tx); err != nil {return err}- return errors.Wrap(tx.Commit(), "failed to commit batch database write transaction")+ return errors.Wrap(tx.Commit(), "failed to commit batch database transaction")}// lockDatabaseBatchTargets locks the batch targets in full primary-key order.-// Dynamic environment queries re-scan the original where clause so rows that-// start matching after the pre-read are detected. Static targets re-apply the-// clause to the pre-read identities, so rows that no longer match are skipped.// Only db rows are locked.-func lockDatabaseBatchTargets(ctx context.Context, tx *sql.Tx, where *qb.Query, targets []databaseBatchTarget, dynamic bool) ([]databaseBatchTarget, error) {- if len(targets) == 0 && !dynamic {- return nil, nil- }+func lockDatabaseBatchTargets(ctx context.Context, tx *sql.Tx, where *qb.Query) ([]databaseBatchTarget, error) {q := qb.Q().Space(`- SELECT db.instance, db.name, db.project, instance.project+ SELECT db.instance, db.name, db.projectFROM dbJOIN instance ON instance.resource_id = db.instanceWHERE ?`, where)- if !dynamic {- instances, names := make([]string, 0, len(targets)), make([]string, 0, len(targets))- for _, target := range targets {- instances, names = append(instances, target.instanceID), append(names, target.databaseName)- }- q.And(`(db.instance, db.name) IN (SELECT * FROM unnest(?::TEXT[], ?::TEXT[]))`, instances, names)- }q.Space("ORDER BY db.instance, db.name FOR UPDATE OF db")query, args, err := q.ToSQL()if err != nil {@@ -678,13 +528,9 @@ func lockDatabaseBatchTargets(ctx context.Context, tx *sql.Tx, where *qb.Query,var locked []databaseBatchTargetfor rows.Next() {var target databaseBatchTarget- var instanceProject sql.NullString- if err := rows.Scan(&target.instanceID, &target.databaseName, &target.projectID, &instanceProject); err != nil {+ if err := rows.Scan(&target.instanceID, &target.databaseName, &target.projectID); err != nil {return nil, errors.Wrap(err, "failed to scan locked batch database target")}- if instanceProject.Valid {- target.instanceProject = instanceProject.String- }locked = append(locked, target)}if err := rows.Err(); err != nil {@@ -758,43 +604,17 @@ type databaseOwnership struct {instanceProject *string}-// withDatabasePurgeFence serializes a database write with direct project and-// instance purge. Database descendants may be absent, so row locks alone cannot-// prevent a purge from passing their table before the writer inserts one.-//-// Database sync intentionally supports an archived (but still existing) project-// owner. It does not support a purged or soft-deleted instance: that state is a-// direct-purge boundary and no new database data may be written through it.-func (s *Store) withDatabasePurgeFence(+func (s *Store) withDatabaseWrite(ctx context.Context,- instanceID, databaseName, requestedProjectID string,+ instanceID, databaseName string,lockChild func(*sql.Tx) error,write func(*sql.Tx, *databaseOwnership) error,) error {- projectID, err := s.databasePurgeProject(ctx, instanceID, databaseName, requestedProjectID)- if err != nil {- return err- }-tx, err := s.GetDB().BeginTx(ctx, nil)if err != nil {return errors.Wrap(err, "failed to begin database write transaction")}defer tx.Rollback()- projectFences := []string{projectID}- if requestedProjectID != "" {- projectFences = append(projectFences, requestedProjectID)- }- slices.Sort(projectFences)- projectFences = slices.Compact(projectFences)- for _, fence := range projectFences {- if err := acquireProjectPurgeLock(ctx, tx, fence); err != nil {- return errors.Wrapf(err, "failed to lock project purge fence for %s", fence)- }- }- if err := acquireInstancePurgeLock(ctx, tx, instanceID); err != nil {- return errors.Wrapf(err, "failed to lock instance purge fence for %s", instanceID)- }if lockChild != nil {if err := lockChild(tx); err != nil {return err@@ -805,54 +625,12 @@ func (s *Store) withDatabasePurgeFence(if err != nil {return err}- if ownership.instanceProject != nil {- projectID = *ownership.instanceProject- } else if requestedProjectID == "" && ownership.exists {- projectID = ownership.projectID- }- if !slices.Contains(projectFences, projectID) {- return errors.Errorf("database ownership changed to project %s; retry", projectID)- }- for _, projectFence := range projectFences {- var foundProjectID string- if err := tx.QueryRowContext(ctx, "SELECT resource_id FROM project WHERE resource_id = $1 FOR UPDATE", projectFence).Scan(&foundProjectID); err != nil {- if errors.Is(err, sql.ErrNoRows) {- return common.Errorf(common.NotFound, "project %s not found", projectFence)- }- return errors.Wrapf(err, "failed to lock project %s", projectFence)- }- }if err := write(tx, ownership); err != nil {return err}return errors.Wrap(tx.Commit(), "failed to commit database write transaction")}-func (s *Store) databasePurgeProject(ctx context.Context, instanceID, databaseName, requestedProjectID string) (string, error) {- var instanceProject, databaseProject sql.NullString- if err := s.GetDB().QueryRowContext(ctx, `- SELECT instance.project, db.project- FROM instance- LEFT JOIN db ON db.instance = instance.resource_id AND db.name = $2- WHERE instance.resource_id = $1- `, instanceID, databaseName).Scan(&instanceProject, &databaseProject); err != nil {- if errors.Is(err, sql.ErrNoRows) {- return "", common.Errorf(common.NotFound, "instance %s not found", instanceID)- }- return "", errors.Wrapf(err, "failed to find database purge project for instance %s", instanceID)- }- if instanceProject.Valid {- return instanceProject.String, nil- }- if databaseProject.Valid {- return databaseProject.String, nil- }- if requestedProjectID != "" {- return requestedProjectID, nil- }- return "", common.Errorf(common.NotFound, "database %s not found", common.FormatDatabase(instanceID, databaseName))-}-func lockDatabaseOwnership(ctx context.Context, tx *sql.Tx, instanceID, databaseName string) (*databaseOwnership, error) {ownership := &databaseOwnership{}var databaseProject sql.NullString@@ -871,12 +649,12 @@ func lockDatabaseOwnership(ctx context.Context, tx *sql.Tx, instanceID, databasevar instanceProject sql.NullStringvar deleted boolif err := tx.QueryRowContext(ctx, `- SELECT project, deleted FROM instance WHERE resource_id = $1 FOR NO KEY UPDATE+ SELECT project, deleted FROM instance WHERE resource_id = $1`, instanceID).Scan(&instanceProject, &deleted); err != nil {if errors.Is(err, sql.ErrNoRows) {return nil, common.Errorf(common.NotFound, "instance %s not found", instanceID)}- return nil, errors.Wrapf(err, "failed to lock instance %s", instanceID)+ return nil, errors.Wrapf(err, "failed to find instance %s", instanceID)}if deleted {return nil, common.Errorf(common.NotFound, "instance %s is deleted", instanceID)diff --git a/backend/store/database_ownership_test.go b/backend/store/database_ownership_test.goindex 7b9f81134d..84f21ac25b 100644--- a/backend/store/database_ownership_test.go+++ b/backend/store/database_ownership_test.go@@ -11,7 +11,7 @@ import ()func TestDatabaseWritersRespectProjectInstanceOwnership(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `+ fixture := newStorePostgresFixture(t, `ALTER TABLE instance ADD COLUMN IF NOT EXISTS project TEXT REFERENCES project(resource_id);INSERT INTO project (resource_id, workspace, name) VALUES('project-b', 'default', 'Project B'),@@ -117,7 +117,7 @@ func TestDatabaseWritersRespectProjectInstanceOwnership(t *testing.T) {}func TestListDatabasesIncludesInstanceProject(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `+ fixture := newStorePostgresFixture(t, `ALTER TABLE instance ADD COLUMN IF NOT EXISTS project TEXT REFERENCES project(resource_id);INSERT INTO project (resource_id, workspace, name) VALUES ('project-b', 'default', 'Project B');INSERT INTO instance (resource_id, workspace, project) VALUES@@ -155,7 +155,7 @@ func getDatabaseProject(ctx context.Context, t *testing.T, s *store.Store, insta}func TestProjectInstanceDatabaseUpsertAllowsArchivedOwner(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `+ fixture := newStorePostgresFixture(t, `ALTER TABLE instance ADD COLUMN IF NOT EXISTS project TEXT REFERENCES project(resource_id);INSERT INTO project (resource_id, workspace, name, deleted)VALUES ('project-b', 'default', 'Project B', TRUE);@@ -174,7 +174,7 @@ func TestProjectInstanceDatabaseUpsertAllowsArchivedOwner(t *testing.T) {}func TestProjectInstanceDatabaseUpdateAllowsArchivedOwner(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `+ fixture := newStorePostgresFixture(t, `ALTER TABLE instance ADD COLUMN IF NOT EXISTS project TEXT REFERENCES project(resource_id);INSERT INTO project (resource_id, workspace, name, deleted)VALUES ('project-b', 'default', 'Project B', TRUE);diff --git a/backend/store/database_purge_writer_test.go b/backend/store/database_purge_writer_test.godeleted file mode 100644index ee333295cb..0000000000--- a/backend/store/database_purge_writer_test.go+++ /dev/null@@ -1,146 +0,0 @@-package store_test--import (- "strings"- "testing"-- "github.com/stretchr/testify/require"-- storepb "github.com/bytebase/bytebase/backend/generated-go/store"-)--func TestCreateSyncHistoryAndDeleteProjectSerialize(t *testing.T) {- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', 'project-a');- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'project-a');- INSERT INTO sync_history (instance, db_name) VALUES ('instance-a', 'db-a');- `- createHistory := func(fixture *projectDeletionLockOrderFixture) error {- _, err := fixture.store.CreateSyncHistory(fixture.ctx, "instance-a", "db-a", &storepb.DatabaseSchemaMetadata{}, "")- return err- }-- t.Run("purge first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9931- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER DELETE ON sync_history FOR EACH ROW")-- purgeResult := make(chan error, 1)- go func() { purgeResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- writerResult := make(chan error, 1)- go func() { writerResult <- createHistory(fixture) }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, purgeResult)- err := <-writerResult- require.Error(t, err)- require.NotContains(t, strings.ToLower(err.Error()), "foreign key")- })-- t.Run("writer first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9932- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- _, err := fixture.db.ExecContext(fixture.ctx, `- CREATE FUNCTION block_sync_history_insert() RETURNS trigger AS $$- BEGIN- PERFORM pg_advisory_xact_lock(9932);- RETURN NEW;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER block_sync_history_insert- BEFORE INSERT ON sync_history FOR EACH ROW- EXECUTE FUNCTION block_sync_history_insert();- `)- require.NoError(t, err)-- writerResult := make(chan error, 1)- go func() { writerResult <- createHistory(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- purgeResult := make(chan error, 1)- go func() { purgeResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, writerResult)- requireMaintenanceResult(t, purgeResult)- })-}--func TestCreateSyncHistoryAndDeleteInstanceSerialize(t *testing.T) {- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', 'project-a');- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'project-a');- INSERT INTO sync_history (instance, db_name) VALUES ('instance-a', 'db-a');- `- createHistory := func(fixture *projectDeletionLockOrderFixture) error {- _, err := fixture.store.CreateSyncHistory(fixture.ctx, "instance-a", "db-a", &storepb.DatabaseSchemaMetadata{}, "")- return err- }- archiveAndPurge := func(fixture *projectDeletionLockOrderFixture) error {- if _, err := fixture.db.ExecContext(fixture.ctx, `- UPDATE instance SET deleted = TRUE WHERE resource_id = 'instance-a'- `); err != nil {- return err- }- return fixture.store.DeleteInstance(fixture.ctx, "default", "instance-a")- }-- t.Run("purge first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9933- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER DELETE ON sync_history FOR EACH ROW")-- purgeResult := make(chan error, 1)- go func() { purgeResult <- archiveAndPurge(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- writerResult := make(chan error, 1)- go func() { writerResult <- createHistory(fixture) }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, purgeResult)- err := <-writerResult- require.Error(t, err)- require.NotContains(t, strings.ToLower(err.Error()), "foreign key")- })-- t.Run("writer first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9934- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- _, err := fixture.db.ExecContext(fixture.ctx, `- CREATE FUNCTION block_sync_history_insert() RETURNS trigger AS $$- BEGIN- PERFORM pg_advisory_xact_lock(9934);- RETURN NEW;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER block_sync_history_insert- BEFORE INSERT ON sync_history FOR EACH ROW- EXECUTE FUNCTION block_sync_history_insert();- `)- require.NoError(t, err)-- writerResult := make(chan error, 1)- go func() { writerResult <- createHistory(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- purgeResult := make(chan error, 1)- go func() { purgeResult <- archiveAndPurge(fixture) }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, writerResult)- requireMaintenanceResult(t, purgeResult)- })-}diff --git a/backend/store/db_schema.go b/backend/store/db_schema.goindex e4cd8fe86c..366354c06c 100644--- a/backend/store/db_schema.go+++ b/backend/store/db_schema.go@@ -118,7 +118,7 @@ func (s *Store) UpsertDBSchema(return errors.Wrapf(err, "failed to build sql")}- err = s.withDatabasePurgeFence(ctx, instanceID, databaseName, "", func(tx *sql.Tx) error {+ err = s.withDatabaseWrite(ctx, instanceID, databaseName, func(tx *sql.Tx) error {_, err := tx.ExecContext(ctx, "SELECT 1 FROM db_schema WHERE instance = $1 AND db_name = $2 FOR UPDATE", instanceID, databaseName)return err}, func(tx *sql.Tx, _ *databaseOwnership) error {@@ -155,7 +155,7 @@ func (s *Store) UpdateDBSchema(ctx context.Context, instanceID, databaseName strreturn errors.Wrapf(err, "failed to build sql")}- err = s.withDatabasePurgeFence(ctx, instanceID, databaseName, "", func(tx *sql.Tx) error {+ err = s.withDatabaseWrite(ctx, instanceID, databaseName, func(tx *sql.Tx) error {_, err := tx.ExecContext(ctx, "SELECT 1 FROM db_schema WHERE instance = $1 AND db_name = $2 FOR UPDATE", instanceID, databaseName)return err}, func(tx *sql.Tx, _ *databaseOwnership) error {diff --git a/backend/store/iam_policy_test.go b/backend/store/iam_policy_test.goindex 86a0019b80..4123baef3b 100644--- a/backend/store/iam_policy_test.go+++ b/backend/store/iam_policy_test.go@@ -48,7 +48,7 @@ func TestSetIamPolicyCompareAndSwap(t *testing.T) {INSERT INTO policy (workspace, resource_type, resource, type, payload, inherit_from_parent)VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx, s := fixture.ctx, fixture.storeresource := common.FormatProject("project-a")@@ -123,7 +123,7 @@ func TestSetIamPolicyComparesAgainstTheRowNotTheCache(t *testing.T) {INSERT INTO policy (workspace, resource_type, resource, type, payload, inherit_from_parent)VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx := fixture.ctxcached, err := store.New(ctx, fixture.pgURL, true)@@ -187,7 +187,7 @@ func TestSetIamPolicyValidatesAgainstTheReplacedPolicy(t *testing.T) {INSERT INTO policy (workspace, resource_type, resource, type, payload, inherit_from_parent)VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx, s := fixture.ctx, fixture.storevar seen []string@@ -222,7 +222,7 @@ func TestSetIamPolicyIsScopedToItsWorkspace(t *testing.T) {VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE),('other', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx, s := fixture.ctx, fixture.store_, _, err := s.SetIamPolicy(ctx, &store.SetIamPolicyMessage{@@ -242,7 +242,7 @@ func TestSetIamPolicyIsScopedToItsWorkspace(t *testing.T) {// A resource whose IAM policy was never written has no row to lock, so the// first write creates one rather than failing the compare.func TestSetIamPolicyCreatesTheFirstPolicy(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, "")+ fixture := newStorePostgresFixture(t, "")ctx, s := fixture.ctx, fixture.storeabsent, err := s.GetProjectIamPolicy(ctx, "default", "project-a")@@ -267,7 +267,7 @@ func TestSetIamPolicyCreatesTheFirstPolicy(t *testing.T) {// policy, and now does it inside the locked transaction. Its merge semantics// must be unchanged by that move.func TestPatchWorkspaceIamPolicyMerges(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, "")+ fixture := newStorePostgresFixture(t, "")ctx, s := fixture.ctx, fixture.store_, err := s.PatchWorkspaceIamPolicy(ctx, &store.PatchIamPolicyMessage{@@ -320,7 +320,7 @@ func TestPatchIamPolicyMergesIntoTheStoredPolicy(t *testing.T) {INSERT INTO policy (workspace, resource_type, resource, type, payload, inherit_from_parent)VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx := fixture.ctxresource := common.FormatProject("project-a")@@ -365,7 +365,7 @@ func TestPatchIamPolicyRollsBackAFailedMutation(t *testing.T) {INSERT INTO policy (workspace, resource_type, resource, type, payload, inherit_from_parent)VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx, s := fixture.ctx, fixture.storeboom := errors.New("boom")@@ -391,7 +391,7 @@ func TestSetIamPolicyDoesNotPublishItsOwnWrite(t *testing.T) {INSERT INTO policy (workspace, resource_type, resource, type, payload, inherit_from_parent)VALUES ('default', 'PROJECT', 'projects/project-a', 'IAM', '` + ownerAndDeveloperPolicy + `', FALSE);`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)ctx := fixture.ctxcached, err := store.New(ctx, fixture.pgURL, true)diff --git a/backend/store/id.go b/backend/store/id.goindex 0b25ccd5de..f402ee42bb 100644--- a/backend/store/id.go+++ b/backend/store/id.go@@ -15,11 +15,24 @@ import (// reserving IDs below 101 for seed/test data.const idMinValue int64 = 101-// lockActiveProject locks the project row and requires it to be active,-// serializing the caller with project deletion. Creation is rejected with-// NotFound when the project is missing or deleted. This is the "requires an-// active project" lifecycle policy from backend/store/AGENTS.md; call it-// after locking any existing child rows.+// requireActiveProject requires the project to exist and be active.+func requireActiveProject(ctx context.Context, tx *sql.Tx, projectID string) error {+ var deleted bool+ if err := tx.QueryRowContext(ctx,+ "SELECT deleted FROM project WHERE resource_id = $1", projectID).Scan(&deleted); err != nil {+ if errors.Is(err, sql.ErrNoRows) {+ return common.Errorf(common.NotFound, "project %s not found", projectID)+ }+ return errors.Wrapf(err, "failed to find project %s", projectID)+ }+ if deleted {+ return common.Errorf(common.NotFound, "project %s is deleted", projectID)+ }+ return nil+}++// lockActiveProject locks the project row and requires it to be active. This+// serializes MAX(id) + 1 allocation for project-scoped tables.func lockActiveProject(ctx context.Context, tx *sql.Tx, projectID string) error {var deleted boolif err := tx.QueryRowContext(ctx,@@ -37,8 +50,7 @@ func lockActiveProject(ctx context.Context, tx *sql.Tx, projectID string) error// nextProjectID returns the next per-project auto-increment ID for the given table.// Must be called within a transaction. Locks and validates the active project row to-// serialize concurrent inserts with project deletion.-// Callers must first lock existing child rows as described in backend/store/AGENTS.md.+// serialize concurrent allocations.// Returns at least idMinValue (101) for new projects.func nextProjectID(ctx context.Context, tx *sql.Tx, table, projectID string) (int64, error) {if err := lockActiveProject(ctx, tx, projectID); err != nil {diff --git a/backend/store/instance.go b/backend/store/instance.goindex 6ea9dcf155..70895645ae 100644--- a/backend/store/instance.go+++ b/backend/store/instance.go@@ -238,12 +238,11 @@ func (s *Store) CreateInstance(ctx context.Context, instanceCreate *InstanceMessSELECT deletedFROM projectWHERE resource_id = $1 AND workspace = $2- FOR UPDATE`, *instanceCreate.ProjectID, instanceCreate.Workspace).Scan(&deleted); err != nil {if errors.Is(err, sql.ErrNoRows) {return nil, common.Errorf(common.NotFound, "project %s not found", *instanceCreate.ProjectID)}- return nil, errors.Wrapf(err, "failed to lock project %s", *instanceCreate.ProjectID)+ return nil, errors.Wrapf(err, "failed to find project %s", *instanceCreate.ProjectID)}if deleted {return nil, common.Errorf(common.NotFound, "project %s is deleted", *instanceCreate.ProjectID)@@ -370,32 +369,11 @@ func (s *Store) UpdateInstance(ctx context.Context, patch *UpdateInstanceMessagefunc (s *Store) updateInstanceLifecycle(ctx context.Context, patch *UpdateInstanceMessage, q *qb.Query) (*InstanceMessage, error) {resourceID := *patch.ResourceID- var projectFences []string- if destinationProjectID := patch.MoveDatabasesToProjectID; destinationProjectID != nil {- var err error- projectFences, err = s.listInstanceDatabaseProjects(ctx, resourceID)- if err != nil {- return nil, err- }- projectFences = append(projectFences, *destinationProjectID)- slices.Sort(projectFences)- projectFences = slices.Compact(projectFences)- }-tx, err := s.GetDB().BeginTx(ctx, nil)if err != nil {return nil, errors.Wrap(err, "failed to begin instance lifecycle transaction")}defer tx.Rollback()- for _, projectID := range projectFences {- if err := acquireProjectPurgeLock(ctx, tx, projectID); err != nil {- return nil, errors.Wrapf(err, "failed to lock project purge fence for %s", projectID)- }- }- if err := acquireInstancePurgeLock(ctx, tx, resourceID); err != nil {- return nil, errors.Wrapf(err, "failed to lock instance lifecycle fence for %s", resourceID)- }-activeTaskRunCount, err := lockActiveTaskRunsForInstance(ctx, tx, resourceID)if err != nil {return nil, err@@ -414,12 +392,6 @@ func (s *Store) updateInstanceLifecycle(ctx context.Context, patch *UpdateInstanif err != nil {return nil, err}- for _, database := range movedDatabases {- if !slices.Contains(projectFences, database.projectID) {- return nil, common.Errorf(common.Conflict, "database ownership changed to project %s for %s; retry", database.projectID, common.FormatDatabase(resourceID, database.name))- }- }-var instanceProject sql.NullStringif err := tx.QueryRowContext(ctx, `SELECT project FROM instance WHERE resource_id = $1 FOR NO KEY UPDATE@@ -480,28 +452,6 @@ func (s *Store) updateInstanceLifecycle(ctx context.Context, patch *UpdateInstanreturn s.GetInstance(ctx, find)}-func (s *Store) listInstanceDatabaseProjects(ctx context.Context, resourceID string) ([]string, error) {- rows, err := s.GetDB().QueryContext(ctx, `- SELECT DISTINCT project FROM db WHERE instance = $1 ORDER BY project- `, resourceID)- if err != nil {- return nil, errors.Wrapf(err, "failed to find database projects for instance %s", resourceID)- }- defer rows.Close()- var projects []string- for rows.Next() {- var projectID string- if err := rows.Scan(&projectID); err != nil {- return nil, errors.Wrap(err, "failed to scan instance database project")- }- projects = append(projects, projectID)- }- if err := rows.Err(); err != nil {- return nil, errors.Wrap(err, "failed to read instance database projects")- }- return projects, nil-}-type instanceDatabaseTarget struct {name stringprojectID string@@ -855,12 +805,7 @@ func (s *Store) DeleteInstance(ctx context.Context, workspace string, resourceIDreturn errors.Wrap(err, "failed to begin transaction")}defer tx.Rollback()- if err := acquireInstancePurgeLock(ctx, tx, resourceID); err != nil {- return errors.Wrapf(err, "failed to lock instance purge fence for %s", resourceID)- }-- // Delete query history before locking database-scoped rows to preserve the- // canonical sibling-branch order.+ // Delete query history before database-scoped rows.q := qb.Q().Space(`DELETE FROM query_historyWHERE database LIKE 'instances/' || ? || '/databases/%'@@ -908,36 +853,6 @@ func (s *Store) DeleteInstance(ctx context.Context, workspace string, resourceIDreturn errors.Wrapf(err, "failed to delete task_run for instance %s", resourceID)}- // Lock tasks in full primary-key order before deleting them.- q = qb.Q().Space(`- SELECT project, id- FROM task- WHERE instance = ?- ORDER BY project, id- FOR UPDATE- `, resourceID)- query, args, err = q.ToSQL()- if err != nil {- return errors.Wrap(err, "failed to build instance task lock query")- }- if err := func() error {- rows, err := tx.QueryContext(ctx, query, args...)- if err != nil {- return errors.Wrapf(err, "failed to lock tasks for instance %s", resourceID)- }- defer rows.Close()- for rows.Next() {- var projectID string- var taskID int64- if err := rows.Scan(&projectID, &taskID); err != nil {- return errors.Wrap(err, "failed to scan locked task")- }- }- return rows.Err()- }(); err != nil {- return err- }-// Delete tasks associated with this instanceq = qb.Q().Space(`DELETE FROM task WHERE instance = ?@@ -1010,16 +925,14 @@ func (s *Store) DeleteInstance(ctx context.Context, workspace string, resourceIDreturn errors.Wrapf(err, "failed to delete databases for instance %s", resourceID)}- // Lock the instance only after all descendant branches. A project owner is- // locked after its instance, matching the canonical child-to-parent order.- var projectID sql.NullString+ // Lock the instance while checking the required deleted state.var deleted boolif err := tx.QueryRowContext(ctx, `- SELECT project, deleted+ SELECT deletedFROM instanceWHERE resource_id = $1 AND workspace = $2FOR UPDATE- `, resourceID, workspace).Scan(&projectID, &deleted); err != nil {+ `, resourceID, workspace).Scan(&deleted); err != nil {if errors.Is(err, sql.ErrNoRows) {return errors.Errorf("instance %s not found or not marked as deleted", resourceID)}@@ -1028,16 +941,7 @@ func (s *Store) DeleteInstance(ctx context.Context, workspace string, resourceIDif !deleted {return errors.Errorf("instance %s not found or not marked as deleted", resourceID)}- if projectID.Valid {- var lockedProjectID string- if err := tx.QueryRowContext(ctx, `- SELECT resource_id FROM project WHERE resource_id = $1 FOR UPDATE- `, projectID.String).Scan(&lockedProjectID); err != nil {- return errors.Wrapf(err, "failed to lock owning project %s", projectID.String)- }- }-- // Delete the soft-deleted instance after locking its owning project.+ // Delete the soft-deleted instance.q = qb.Q().Space(`DELETE FROM instanceWHERE resource_id = ? AND deleted = TRUE AND workspace = ?diff --git a/backend/store/instance_archive_task_run_test.go b/backend/store/instance_archive_task_run_test.godeleted file mode 100644index d2409e4212..0000000000--- a/backend/store/instance_archive_task_run_test.go+++ /dev/null@@ -1,236 +0,0 @@-package store_test--import (- "strings"- "testing"-- "github.com/stretchr/testify/require"-- "github.com/bytebase/bytebase/backend/common"- "github.com/bytebase/bytebase/backend/store"-)--const instanceArchiveTaskRunSeed = `- UPDATE project SET deleted = FALSE WHERE resource_id = 'project-a';- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', 'project-a');- INSERT INTO plan (id, creator, project, name, description)- VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO task (id, project, plan_id, instance, type)- VALUES (101, 'project-a', 101, 'instance-a', 'DATABASE_CREATE');-`--func archiveInstanceA(fixture *projectDeletionLockOrderFixture) error {- deleted := true- _, err := fixture.store.UpdateInstance(fixture.ctx, &store.UpdateInstanceMessage{- ResourceID: new("instance-a"),- Workspace: "default",- Deleted: &deleted,- })- return err-}--func createInstanceATaskRun(fixture *projectDeletionLockOrderFixture) error {- return fixture.store.CreatePendingTaskRuns(fixture.ctx, "[email protected]", &store.TaskRunMessage{- ProjectID: "project-a",- TaskUID: 101,- })-}--func forceArchiveWorkspaceInstanceA(fixture *projectDeletionLockOrderFixture) error {- deleted := true- destinationProjectID := "default"- _, err := fixture.store.UpdateInstance(fixture.ctx, &store.UpdateInstanceMessage{- ResourceID: new("instance-a"),- Workspace: "default",- Deleted: &deleted,- MoveDatabasesToProjectID: &destinationProjectID,- })- return err-}--func createInstanceADatabase(fixture *projectDeletionLockOrderFixture) error {- _, err := fixture.store.CreateDatabaseDefault(fixture.ctx, &store.DatabaseMessage{- InstanceID: "instance-a",- DatabaseName: "db-a",- ProjectID: "project-a",- })- return err-}--func TestArchiveInstanceAndCreateTaskRunSerialize(t *testing.T) {- t.Run("archive first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, instanceArchiveTaskRunSeed)- const barrierID = 9945- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER UPDATE OF deleted ON instance FOR EACH ROW")-- archiveResult := make(chan error, 1)- go func() { archiveResult <- archiveInstanceA(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- createResult := make(chan error, 1)- go func() { createResult <- createInstanceATaskRun(fixture) }()- barrierReleased := false- defer func() {- if !barrierReleased {- barrier.release(t)- <-archiveResult- <-createResult- }- }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)- barrierReleased = true-- requireMaintenanceResult(t, archiveResult)- createErr := <-createResult- require.Error(t, createErr)- require.Equal(t, common.Conflict, common.ErrorCode(createErr))-- var deleted bool- var activeTaskRunCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT deleted FROM instance WHERE resource_id = 'instance-a'",- ).Scan(&deleted))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, `- SELECT COUNT(*) FROM task_run- WHERE project = 'project-a' AND status IN ('PENDING', 'AVAILABLE', 'RUNNING')- `).Scan(&activeTaskRunCount))- require.True(t, deleted)- require.Zero(t, activeTaskRunCount)- })-- t.Run("creation first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, instanceArchiveTaskRunSeed)- const barrierID = 9946- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER INSERT ON task_run FOR EACH ROW")-- createResult := make(chan error, 1)- go func() { createResult <- createInstanceATaskRun(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- archiveResult := make(chan error, 1)- go func() { archiveResult <- archiveInstanceA(fixture) }()- barrierReleased := false- defer func() {- if !barrierReleased {- barrier.release(t)- <-createResult- <-archiveResult- }- }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)- barrierReleased = true-- requireMaintenanceResult(t, createResult)- archiveErr := <-archiveResult- require.Error(t, archiveErr)- require.Equal(t, common.Conflict, common.ErrorCode(archiveErr))-- var deleted bool- var activeTaskRunCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT deleted FROM instance WHERE resource_id = 'instance-a'",- ).Scan(&deleted))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, `- SELECT COUNT(*) FROM task_run- WHERE project = 'project-a' AND status IN ('PENDING', 'AVAILABLE', 'RUNNING')- `).Scan(&activeTaskRunCount))- require.False(t, deleted)- require.Equal(t, 1, activeTaskRunCount)- })-}--func TestForceArchiveInstanceAndCreateFirstDatabaseSerialize(t *testing.T) {- const seedSQL = `- UPDATE project SET deleted = FALSE WHERE resource_id = 'project-a';- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', NULL);- `-- t.Run("archive first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9947- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER UPDATE OF deleted ON instance FOR EACH ROW")-- archiveResult := make(chan error, 1)- go func() { archiveResult <- forceArchiveWorkspaceInstanceA(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- createResult := make(chan error, 1)- go func() { createResult <- createInstanceADatabase(fixture) }()- barrierReleased := false- defer func() {- if !barrierReleased {- barrier.release(t)- <-archiveResult- <-createResult- }- }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)- barrierReleased = true-- requireMaintenanceResult(t, archiveResult)- createErr := <-createResult- require.Error(t, createErr)- require.NotContains(t, strings.ToLower(createErr.Error()), "foreign key")-- var deleted bool- var databaseCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT deleted FROM instance WHERE resource_id = 'instance-a'",- ).Scan(&deleted))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM db WHERE instance = 'instance-a' AND name = 'db-a'",- ).Scan(&databaseCount))- require.True(t, deleted)- require.Zero(t, databaseCount)- })-- t.Run("creation first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9948- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER INSERT ON db FOR EACH ROW")-- createResult := make(chan error, 1)- go func() { createResult <- createInstanceADatabase(fixture) }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- archiveResult := make(chan error, 1)- go func() { archiveResult <- forceArchiveWorkspaceInstanceA(fixture) }()- barrierReleased := false- defer func() {- if !barrierReleased {- barrier.release(t)- <-createResult- <-archiveResult- }- }()- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)- barrierReleased = true-- requireMaintenanceResult(t, createResult)- archiveErr := <-archiveResult- require.Error(t, archiveErr)- require.Equal(t, common.Conflict, common.ErrorCode(archiveErr))- require.NotContains(t, strings.ToLower(archiveErr.Error()), "foreign key")-- var deleted bool- var projectID string- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT deleted FROM instance WHERE resource_id = 'instance-a'",- ).Scan(&deleted))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT project FROM db WHERE instance = 'instance-a' AND name = 'db-a'",- ).Scan(&projectID))- require.False(t, deleted)- require.Equal(t, "project-a", projectID)- })-}diff --git a/backend/store/instance_purge_lock_order_test.go b/backend/store/instance_purge_lock_order_test.godeleted file mode 100644index f024679f09..0000000000--- a/backend/store/instance_purge_lock_order_test.go+++ /dev/null@@ -1,115 +0,0 @@-package store_test--import (- "context"- "database/sql"- "testing"- "time"-- "github.com/stretchr/testify/require"-)--func waitForBackendBlockedBy(ctx context.Context, t *testing.T, db *sql.DB, blockingPID, waitingPID int) {- t.Helper()- require.Eventually(t, func() bool {- var blocked bool- err := db.QueryRowContext(ctx, `- SELECT $1 = ANY(pg_blocking_pids($2))- `, blockingPID, waitingPID).Scan(&blocked)- return err == nil && blocked- }, maintenanceLockWait, 10*time.Millisecond, "backend %d should wait for backend %d", waitingPID, blockingPID)-}--func TestDeleteInstanceDatabaseLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project, deleted)- VALUES ('instance-a', 'default', 'project-a', TRUE);- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'project-a');- `- testPurgeDatabaseLockOrder(t, seedSQL, func(fixture *projectDeletionLockOrderFixture) error {- return fixture.store.DeleteInstance(fixture.ctx, "default", "instance-a")- }, func(t *testing.T, fixture *projectDeletionLockOrderFixture) {- var projectCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, "SELECT COUNT(*) FROM project WHERE resource_id = 'project-a'").Scan(&projectCount))- require.Equal(t, 1, projectCount, "instance purge must preserve its project owner")- })-}--func TestDeleteProjectDatabaseLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project, deleted)- VALUES ('instance-a', 'default', 'project-a', TRUE);- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'project-a');- `- testPurgeDatabaseLockOrder(t, seedSQL, func(fixture *projectDeletionLockOrderFixture) error {- return fixture.store.DeleteProject(fixture.ctx, "default", "project-a")- }, func(t *testing.T, fixture *projectDeletionLockOrderFixture) {- var projectCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, "SELECT COUNT(*) FROM project WHERE resource_id = 'project-a'").Scan(&projectCount))- require.Zero(t, projectCount, "project purge must remove its project owner")- })-}--func testPurgeDatabaseLockOrder(- t *testing.T,- seedSQL string,- purge func(*projectDeletionLockOrderFixture) error,- assertTerminalState func(*testing.T, *projectDeletionLockOrderFixture),-) {- t.Run("purge locks database before instance", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- instanceTx, err := fixture.db.BeginTx(fixture.ctx, nil)- require.NoError(t, err)- defer instanceTx.Rollback()-- var instanceLockPID int- require.NoError(t, instanceTx.QueryRowContext(fixture.ctx, "SELECT pg_backend_pid()").Scan(&instanceLockPID))- var instanceID string- require.NoError(t, instanceTx.QueryRowContext(fixture.ctx, `- SELECT resource_id FROM instance WHERE resource_id = 'instance-a' FOR UPDATE- `).Scan(&instanceID))-- purgeResult := make(chan error, 1)- go func() { purgeResult <- purge(fixture) }()- var purgePID int- require.Eventually(t, func() bool {- return fixture.db.QueryRowContext(fixture.ctx, `- SELECT COALESCE((- SELECT pid FROM pg_stat_activity- WHERE $1 = ANY(pg_blocking_pids(pid))- ORDER BY pid LIMIT 1- ), 0)- `, instanceLockPID).Scan(&purgePID) == nil && purgePID != 0- }, maintenanceLockWait, 10*time.Millisecond, "purge should wait for the instance lock")-- writerConn, err := fixture.db.Conn(fixture.ctx)- require.NoError(t, err)- defer writerConn.Close()- writerTx, err := writerConn.BeginTx(fixture.ctx, nil)- require.NoError(t, err)- defer writerTx.Rollback()- var writerPID int- require.NoError(t, writerTx.QueryRowContext(fixture.ctx, "SELECT pg_backend_pid()").Scan(&writerPID))- writerResult := make(chan error, 1)- go func() {- var name string- writerResult <- writerTx.QueryRowContext(fixture.ctx, `- SELECT name FROM db- WHERE instance = 'instance-a' AND name = 'db-a'- FOR UPDATE- `).Scan(&name)- }()- waitForBackendBlockedBy(fixture.ctx, t, fixture.db, purgePID, writerPID)-- require.NoError(t, instanceTx.Commit())- requireMaintenanceResult(t, purgeResult)- require.ErrorIs(t, <-writerResult, sql.ErrNoRows)-- var databaseCount, instanceCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, "SELECT COUNT(*) FROM db WHERE instance = 'instance-a'").Scan(&databaseCount))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, "SELECT COUNT(*) FROM instance WHERE resource_id = 'instance-a'").Scan(&instanceCount))- require.Zero(t, databaseCount, "the database should be removed after the waiting writer resumes")- require.Zero(t, instanceCount, "the instance should be removed after the waiting writer resumes")- assertTerminalState(t, fixture)- })-}diff --git a/backend/store/issue_lock_order_test.go b/backend/store/issue_lock_order_test.godeleted file mode 100644index b9292b542a..0000000000--- a/backend/store/issue_lock_order_test.go+++ /dev/null@@ -1,50 +0,0 @@-package store_test--import (- "context"- "database/sql"- "testing"-- "github.com/stretchr/testify/require"-- "github.com/bytebase/bytebase/backend/common"- storepb "github.com/bytebase/bytebase/backend/generated-go/store"- "github.com/bytebase/bytebase/backend/store"-)--func TestCreateIssueDoesNotDeadlockWithProjectDeletion(t *testing.T) {- err := runWithConcurrentProjectDeletion(t, `- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- `, "plan", 9903, func(ctx context.Context, s *store.Store) error {- planUID := int64(101)- _, err := s.CreateIssue(ctx, &store.IssueMessage{- ProjectID: "project-a",- CreatorEmail: "[email protected]",- Title: "Issue A",- Type: storepb.Issue_DATABASE_CHANGE,- Payload: &storepb.Issue{},- PlanUID: &planUID,- })- return err- })- require.ErrorIs(t, err, sql.ErrNoRows)-}--func TestCreateIssueRejectsDeletedProjectDuringProjectDeletion(t *testing.T) {- operationErr, deleteErr := runWithCreationBeforeProjectDeletion(t, `- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- `, "plan", func(ctx context.Context, s *store.Store) error {- planUID := int64(101)- _, err := s.CreateIssue(ctx, &store.IssueMessage{- ProjectID: "project-a",- CreatorEmail: "[email protected]",- Title: "Issue A",- Type: storepb.Issue_DATABASE_CHANGE,- Payload: &storepb.Issue{},- PlanUID: &planUID,- })- return err- })- require.NoError(t, deleteErr)- require.Equal(t, common.NotFound, common.ErrorCode(operationErr))-}diff --git a/backend/store/lock_order_test.go b/backend/store/lock_order_test.goindex b20c15b37e..65c0c8aa77 100644--- a/backend/store/lock_order_test.go+++ b/backend/store/lock_order_test.go@@ -14,7 +14,7 @@ import ("github.com/bytebase/bytebase/backend/store")-type projectDeletionLockOrderFixture struct {+type storePostgresFixture struct {ctx context.Contextdb *sql.DBstore *store.Store@@ -23,7 +23,7 @@ type projectDeletionLockOrderFixture struct {pgURL string}-func newProjectDeletionLockOrderFixture(t *testing.T, seedSQL string) *projectDeletionLockOrderFixture {+func newStorePostgresFixture(t *testing.T, seedSQL string) *storePostgresFixture {t.Helper()ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)t.Cleanup(cancel)@@ -47,163 +47,91 @@ func newProjectDeletionLockOrderFixture(t *testing.T, seedSQL string) *projectDerequire.NoError(t, err)t.Cleanup(func() { require.NoError(t, s.Close()) })- return &projectDeletionLockOrderFixture{ctx: ctx, db: db, store: s, pgURL: pgURL}+ return &storePostgresFixture{ctx: ctx, db: db, store: s, pgURL: pgURL}}-func runWithConcurrentProjectDeletion(- t *testing.T,- seedSQL string,- blockedTable string,- advisoryLockID int,- operation func(context.Context, *store.Store) error,-) error {- t.Helper()- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- ctx, db, s := fixture.ctx, fixture.db, fixture.store+const maintenanceLockWait = 10 * time.Second++type maintenanceLockBarrier struct {+ ctx context.Context+ conn *sql.Conn+ id int+}- lockConn, err := db.Conn(ctx)+func newMaintenanceLockBarrier(ctx context.Context, t *testing.T, db *sql.DB, id int) *maintenanceLockBarrier {+ t.Helper()+ conn, err := db.Conn(ctx)require.NoError(t, err)- defer lockConn.Close()- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", advisoryLockID)+ _, err = conn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", id)require.NoError(t, err)- lockReleased := false- defer func() {- if !lockReleased {- _, _ = lockConn.ExecContext(context.Background(), "SELECT pg_advisory_unlock($1)", advisoryLockID)- }- }()- _, err = db.ExecContext(ctx, fmt.Sprintf(`- CREATE FUNCTION block_%[1]s_delete() RETURNS trigger AS $$+ barrier := &maintenanceLockBarrier{ctx: ctx, conn: conn, id: id}+ t.Cleanup(func() { barrier.release(t) })+ return barrier+}++func (b *maintenanceLockBarrier) release(t *testing.T) {+ t.Helper()+ if b.conn == nil {+ return+ }+ _, err := b.conn.ExecContext(b.ctx, "SELECT pg_advisory_unlock($1)", b.id)+ require.NoError(t, err)+ require.NoError(t, b.conn.Close())+ b.conn = nil+}++func installMaintenanceLockBarrier(t *testing.T, db *sql.DB, id int, trigger string) {+ t.Helper()+ _, err := db.Exec(fmt.Sprintf(`+ CREATE FUNCTION maintenance_lock_barrier_%[1]d() RETURNS trigger AS $$BEGIN- PERFORM pg_advisory_xact_lock(%[2]d);- RETURN OLD;+ PERFORM pg_advisory_xact_lock(%[1]d);+ RETURN NEW;END;$$ LANGUAGE plpgsql;- CREATE TRIGGER block_%[1]s_delete- AFTER DELETE ON %[1]s- FOR EACH ROW EXECUTE FUNCTION block_%[1]s_delete();- `, blockedTable, advisoryLockID))+ CREATE TRIGGER maintenance_lock_barrier_%[1]d+ %[2]s EXECUTE FUNCTION maintenance_lock_barrier_%[1]d();+ `, id, trigger))require.NoError(t, err)+}- deleteResult := make(chan error, 1)- go func() {- deleteResult <- s.DeleteProject(ctx, "default", "project-a")- }()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- )- `, advisoryLockID).Scan(&waiting)- return err == nil && waiting- }, 10*time.Second, 10*time.Millisecond, "project deletion should reach the blocked %s delete", blockedTable)-- operationResult := make(chan error, 1)- go func() {- operationResult <- operation(ctx, s)- }()+func waitForMaintenanceBarrier(ctx context.Context, t *testing.T, db *sql.DB, id int) {+ t.Helper()require.Eventually(t, func() bool {var waiting bool- err := db.QueryRowContext(ctx, `- WITH delete_backend AS (- SELECT pid- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- LIMIT 1- )- SELECT EXISTS (- SELECT 1- FROM pg_stat_activity AS activity, delete_backend- WHERE delete_backend.pid = ANY(pg_blocking_pids(activity.pid))- )- `, advisoryLockID).Scan(&waiting)+ err := db.QueryRowContext(ctx, `SELECT EXISTS (+ SELECT 1 FROM pg_locks WHERE locktype = 'advisory' AND objid = $1 AND NOT granted+ )`, id).Scan(&waiting)return err == nil && waiting- }, 10*time.Second, 10*time.Millisecond, "the competing operation should wait behind project deletion")-- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_unlock($1)", advisoryLockID)- require.NoError(t, err)- lockReleased = true- deleteErr := <-deleteResult- operationErr := <-operationResult- require.NoError(t, deleteErr)- return operationErr+ }, maintenanceLockWait, 10*time.Millisecond)}-func runWithCreationBeforeProjectDeletion(- t *testing.T,- seedSQL string,- lockedChildTable string,- operation func(context.Context, *store.Store) error,-) (error, error) {+func maintenanceBarrierWaitingPID(ctx context.Context, t *testing.T, db *sql.DB, id int) int {t.Helper()- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- ctx, db, s := fixture.ctx, fixture.db, fixture.store+ var pid int+ require.NoError(t, db.QueryRowContext(ctx, `+ SELECT pid FROM pg_locks WHERE locktype = 'advisory' AND objid = $1 AND NOT granted LIMIT 1+ `, id).Scan(&pid))+ return pid+}- lockConn, err := db.Conn(ctx)- require.NoError(t, err)- defer lockConn.Close()- lockTx, err := lockConn.BeginTx(ctx, nil)- require.NoError(t, err)- lockReleased := false- defer func() {- if !lockReleased {- _ = lockTx.Rollback()- }- }()-- var lockBackendPID int- require.NoError(t, lockTx.QueryRowContext(ctx, "SELECT pg_backend_pid()").Scan(&lockBackendPID))- var lockedProjectID string- require.NoError(t, lockTx.QueryRowContext(ctx, `- SELECT resource_id- FROM project- WHERE resource_id = 'project-a'- FOR UPDATE- `).Scan(&lockedProjectID))- require.Equal(t, "project-a", lockedProjectID)-- operationResult := make(chan error, 1)- go func() {- operationResult <- operation(ctx, s)- }()-- var operationBackendPID int- require.Eventually(t, func() bool {- return db.QueryRowContext(ctx, `- SELECT COALESCE((- SELECT pid- FROM pg_stat_activity- WHERE $1 = ANY(pg_blocking_pids(pid))- ORDER BY pid- LIMIT 1- ), 0)- `, lockBackendPID).Scan(&operationBackendPID) == nil && operationBackendPID != 0- }, 10*time.Second, 10*time.Millisecond, "creation should lock %s, then wait for the project row", lockedChildTable)-- deleteResult := make(chan error, 1)- go func() {- deleteResult <- s.DeleteProject(ctx, "default", "project-a")- }()+func waitForBackendBlockedByPID(ctx context.Context, t *testing.T, db *sql.DB, pid int) {+ t.Helper()require.Eventually(t, func() bool {var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1- FROM pg_stat_activity- WHERE $1 = ANY(pg_blocking_pids(pid))- )- `, operationBackendPID).Scan(&waiting)+ err := db.QueryRowContext(ctx, `SELECT EXISTS (+ SELECT 1 FROM pg_stat_activity WHERE $1 = ANY(pg_blocking_pids(pid))+ )`, pid).Scan(&waiting)return err == nil && waiting- }, 10*time.Second, 10*time.Millisecond, "project deletion should pass empty child cleanup, then wait for the locked %s", lockedChildTable)+ }, maintenanceLockWait, 10*time.Millisecond)+}- require.NoError(t, lockTx.Commit())- lockReleased = true- return <-operationResult, <-deleteResult+func requireMaintenanceResult(t *testing.T, result <-chan error) {+ t.Helper()+ select {+ case err := <-result:+ require.NoError(t, err)+ case <-time.After(maintenanceLockWait):+ t.Fatal("timed out waiting for maintenance operation")+ }}diff --git a/backend/store/maintenance_lock_order_test.go b/backend/store/maintenance_lock_order_test.godeleted file mode 100644index cf562ff5b1..0000000000--- a/backend/store/maintenance_lock_order_test.go+++ /dev/null@@ -1,434 +0,0 @@-package store_test--import (- "context"- "database/sql"- "fmt"- "testing"- "time"-- "github.com/stretchr/testify/require"-- "github.com/bytebase/bytebase/backend/store"-)--const maintenanceLockWait = 10 * time.Second--type maintenanceLockBarrier struct {- ctx context.Context- conn *sql.Conn- id int-}--func newMaintenanceLockBarrier(ctx context.Context, t *testing.T, db *sql.DB, id int) *maintenanceLockBarrier {- t.Helper()- conn, err := db.Conn(ctx)- require.NoError(t, err)- _, err = conn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", id)- require.NoError(t, err)-- barrier := &maintenanceLockBarrier{ctx: ctx, conn: conn, id: id}- t.Cleanup(func() {- _, _ = conn.ExecContext(context.Background(), "SELECT pg_advisory_unlock($1)", id)- require.NoError(t, conn.Close())- })- return barrier-}--func (b *maintenanceLockBarrier) release(t *testing.T) {- t.Helper()- _, err := b.conn.ExecContext(b.ctx, "SELECT pg_advisory_unlock($1)", b.id)- require.NoError(t, err)-}--func installMaintenanceLockBarrier(t *testing.T, db *sql.DB, id int, trigger string) {- t.Helper()- _, err := db.ExecContext(context.Background(), fmt.Sprintf(`- CREATE FUNCTION maintenance_lock_barrier_%[1]d() RETURNS trigger AS $$- BEGIN- PERFORM pg_advisory_xact_lock(%[1]d);- RETURN NULL;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER maintenance_lock_barrier_%[1]d- %[2]s EXECUTE FUNCTION maintenance_lock_barrier_%[1]d();- `, id, trigger))- require.NoError(t, err)-}--func waitForMaintenanceBarrier(ctx context.Context, t *testing.T, db *sql.DB, id int) {- t.Helper()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- )- `, id).Scan(&waiting)- return err == nil && waiting- }, maintenanceLockWait, 10*time.Millisecond, "operation should reach the advisory barrier")-}--func maintenanceBarrierWaitingPID(ctx context.Context, t *testing.T, db *sql.DB, id int) int {- t.Helper()- var pid int- require.NoError(t, db.QueryRowContext(ctx, `- SELECT pid- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- ORDER BY pid- LIMIT 1- `, id).Scan(&pid))- return pid-}--func waitForBackendBlockedByPID(ctx context.Context, t *testing.T, db *sql.DB, blockingPID int) {- t.Helper()- var waitingPID int- require.Eventually(t, func() bool {- return db.QueryRowContext(ctx, `- SELECT COALESCE((- SELECT pid- FROM pg_stat_activity- WHERE $1 = ANY(pg_blocking_pids(pid))- ORDER BY pid- LIMIT 1- ), 0)- `, blockingPID).Scan(&waitingPID) == nil && waitingPID != 0- }, maintenanceLockWait, 10*time.Millisecond, "operation should wait for backend %d", blockingPID)-}--func waitForOperationBehindMaintenanceBarrier(ctx context.Context, t *testing.T, db *sql.DB, id int) {- t.Helper()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- WITH barrier_backend AS (- SELECT pid- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- LIMIT 1- )- SELECT EXISTS (- SELECT 1- FROM pg_stat_activity AS activity, barrier_backend- WHERE barrier_backend.pid = ANY(pg_blocking_pids(activity.pid))- )- `, id).Scan(&waiting)- return err == nil && waiting- }, maintenanceLockWait, 10*time.Millisecond, "competing operation should wait behind the barrier owner")-}--func requireMaintenanceResult(t *testing.T, result <-chan error) {- t.Helper()- select {- case err := <-result:- require.NoError(t, err)- case <-time.After(maintenanceLockWait):- t.Fatal("timed out waiting for maintenance operation")- }-}--func requireProjectInstanceDeletionState(ctx context.Context, t *testing.T, db *sql.DB) {- t.Helper()- for _, table := range []string{"project", "instance", "db", "revision", "service_account"} {- var count int- require.NoError(t, db.QueryRowContext(ctx, fmt.Sprintf("SELECT COUNT(*) FROM %s", table)).Scan(&count))- if table == "project" {- require.Equal(t, 1, count, "only the default project should remain")- } else {- require.Zero(t, count, "%s should be removed", table)- }- }-- // The connected-database reference now lives in the payload as a soft- // canonical name that simply dangles after instance deletion, so only- // the purge re-parenting is asserted here.- var project string- require.NoError(t, db.QueryRowContext(ctx, `- SELECT project- FROM saved_query- WHERE resource_id = 'saved-query-a'- `).Scan(&project))- require.Equal(t, "default", project)-}--func requireProjectUserDeletionState(ctx context.Context, t *testing.T, db *sql.DB) {- t.Helper()- var projectCount, planCount, queryHistoryCount int- require.NoError(t, db.QueryRowContext(ctx, "SELECT COUNT(*) FROM project").Scan(&projectCount))- require.NoError(t, db.QueryRowContext(ctx, "SELECT COUNT(*) FROM plan").Scan(&planCount))- require.NoError(t, db.QueryRowContext(ctx, "SELECT COUNT(*) FROM query_history").Scan(&queryHistoryCount))- require.Equal(t, 1, projectCount, "only the default project should remain")- require.Zero(t, planCount)- require.Zero(t, queryHistoryCount)-- var email string- require.NoError(t, db.QueryRowContext(ctx, "SELECT email FROM principal WHERE id = 101").Scan(&email))- require.Equal(t, "[email protected]", email)-}--func requireUserInstanceDeletionState(ctx context.Context, t *testing.T, db *sql.DB) {- t.Helper()- for _, table := range []string{"instance", "db", "query_history", "task_run", "task"} {- var count int- require.NoError(t, db.QueryRowContext(ctx, fmt.Sprintf("SELECT COUNT(*) FROM %s", table)).Scan(&count))- require.Zero(t, count, "%s should be removed", table)- }-- var email string- require.NoError(t, db.QueryRowContext(ctx, "SELECT email FROM principal WHERE id = 101").Scan(&email))- require.Equal(t, "[email protected]", email)-}--func TestDeleteProjectAndDeleteInstanceLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO instance (resource_id, workspace, project, deleted)- VALUES ('instance-a', 'default', 'project-a', TRUE);- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'project-a');- INSERT INTO service_account (name, email, workspace, service_key_hash, project)- VALUES ('service account', '[email protected]', 'default', 'unused', 'project-a');- INSERT INTO revision (resource_id, instance, db_name, deleter, version)- VALUES ('revision-a', 'instance-a', 'db-a', '[email protected]', 'v1');- INSERT INTO plan (id, creator, project, name, description)- VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO task (id, project, plan_id, instance, type)- VALUES (101, 'project-a', 101, 'instance-a', 'DATABASE_SCHEMA_UPDATE');- INSERT INTO saved_query (resource_id, creator, project, name, statement)- VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '');- `-- t.Run("delete project first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9921- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- // Park the purge on its task delete: `task` is the table both paths- // still contend on. (The purge used to be parked on the saved-query- // re-parent, with DeleteInstance blocking via its saved-query- // nullify UPDATE — that cross-path statement was removed when the- // connected-database reference became a soft payload link.)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER DELETE ON task FOR EACH ROW")-- projectResult := make(chan error, 1)- go func() { projectResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- projectPID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- instanceResult := make(chan error, 1)- go func() { instanceResult <- fixture.store.DeleteInstance(fixture.ctx, "default", "instance-a") }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, projectPID)- barrier.release(t)-- requireMaintenanceResult(t, projectResult)- require.ErrorContains(t, <-instanceResult, "not found or not marked as deleted")- requireProjectInstanceDeletionState(fixture.ctx, t, fixture.db)- })-- t.Run("delete instance first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9922- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER DELETE ON task FOR EACH ROW")-- instanceResult := make(chan error, 1)- go func() { instanceResult <- fixture.store.DeleteInstance(fixture.ctx, "default", "instance-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- instancePID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- projectResult := make(chan error, 1)- go func() { projectResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, instancePID)- barrier.release(t)-- requireMaintenanceResult(t, instanceResult)- requireMaintenanceResult(t, projectResult)- requireProjectInstanceDeletionState(fixture.ctx, t, fixture.db)- })-}--func TestCreateProjectInstanceAndDeleteProjectLockOrder(t *testing.T) {- createProjectInstance := func(fixture *projectDeletionLockOrderFixture) error {- projectID := "project-a"- _, err := fixture.store.CreateInstance(fixture.ctx, &store.InstanceMessage{- ResourceID: "instance-a",- Workspace: "default",- ProjectID: &projectID,- Metadata: testInstanceMetadata(),- })- return err- }-- requireNoProjectInstance := func(t *testing.T, fixture *projectDeletionLockOrderFixture) {- t.Helper()- var projectCount, instanceCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, "SELECT COUNT(*) FROM project").Scan(&projectCount))- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx, "SELECT COUNT(*) FROM instance WHERE resource_id = 'instance-a'").Scan(&instanceCount))- require.Equal(t, 1, projectCount, "only the default project should remain")- require.Zero(t, instanceCount, "the project instance should not survive project deletion")- }-- t.Run("delete project first rejects an absent project instance", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `- INSERT INTO saved_query (resource_id, creator, project, name, statement)- VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '');- `)- const barrierID = 9927- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER UPDATE OF project ON saved_query FOR EACH ROW")-- deleteResult := make(chan error, 1)- go func() { deleteResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- createResult := make(chan error, 1)- go func() { createResult <- createProjectInstance(fixture) }()- select {- case err := <-createResult:- require.Error(t, err)- case <-time.After(maintenanceLockWait):- t.Fatal("project instance creation should reject a deleted project before purge completion")- }- barrier.release(t)-- requireMaintenanceResult(t, deleteResult)- requireNoProjectInstance(t, fixture)- })-- t.Run("new project instance is removed by project deletion", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, "")- _, err := fixture.db.ExecContext(fixture.ctx, `- INSERT INTO instance (resource_id, workspace, project)- VALUES ('instance-a', 'default', 'project-a')- `)- require.NoError(t, err)- require.NoError(t, fixture.store.DeleteProject(fixture.ctx, "default", "project-a"))- requireNoProjectInstance(t, fixture)- })-}--func TestUpdateUserEmailAndDeleteProjectLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO principal (name, email, password_hash) VALUES ('User A', '[email protected]', 'unused');- INSERT INTO plan (id, creator, project, name, description)- VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO query_history (resource_id, creator, project, database, statement, type)- VALUES ('query-history-a', '[email protected]', 'project-a', 'instances/instance-a/databases/db-a', '', 'QUERY');- `-- run := func(t *testing.T, first string) {- t.Helper()- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- user, err := fixture.store.GetUserByEmail(fixture.ctx, "[email protected]")- require.NoError(t, err)- require.NotNil(t, user)-- barrierID := 9923- trigger := "AFTER DELETE ON query_history FOR EACH ROW"- if first == "user" {- barrierID = 9924- trigger = "AFTER UPDATE OF creator ON plan FOR EACH ROW"- }- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, trigger)-- updateUser := func() error {- _, err := fixture.store.UpdateUserEmail(fixture.ctx, user, "[email protected]")- return err- }- deleteProject := func() error { return fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }-- var firstResult, secondResult chan error- if first == "project" {- firstResult, secondResult = make(chan error, 1), make(chan error, 1)- go func() { firstResult <- deleteProject() }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- go func() { secondResult <- updateUser() }()- } else {- firstResult, secondResult = make(chan error, 1), make(chan error, 1)- go func() { firstResult <- updateUser() }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- go func() { secondResult <- deleteProject() }()- }- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, firstResult)- requireMaintenanceResult(t, secondResult)- requireProjectUserDeletionState(fixture.ctx, t, fixture.db)- }-- t.Run("delete project first", func(t *testing.T) { run(t, "project") })- t.Run("update user first", func(t *testing.T) { run(t, "user") })-}--func TestUpdateUserEmailAndDeleteInstanceLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO principal (name, email, password_hash) VALUES ('User A', '[email protected]', 'unused');- INSERT INTO instance (resource_id, workspace, deleted) VALUES ('instance-a', 'default', TRUE);- INSERT INTO db (instance, name, project) VALUES ('instance-a', 'db-a', 'project-a');- INSERT INTO plan (id, creator, project, name, description)- VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO task (id, project, plan_id, instance, type)- VALUES (101, 'project-a', 101, 'instance-a', 'DATABASE_SCHEMA_UPDATE');- INSERT INTO task_run (id, creator, project, task_id, attempt, status)- VALUES (101, '[email protected]', 'project-a', 101, 0, 'DONE');- INSERT INTO query_history (resource_id, creator, project, database, statement, type)- VALUES ('query-history-a', '[email protected]', 'project-a', 'instances/instance-a/databases/db-a', '', 'QUERY');- `-- run := func(t *testing.T, first string) {- t.Helper()- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- user, err := fixture.store.GetUserByEmail(fixture.ctx, "[email protected]")- require.NoError(t, err)- require.NotNil(t, user)-- barrierID := 9925- trigger := "AFTER DELETE ON query_history FOR EACH ROW"- if first == "user" {- barrierID = 9926- trigger = "AFTER UPDATE OF creator ON task_run FOR EACH ROW"- }- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID, trigger)-- updateUser := func() error {- _, err := fixture.store.UpdateUserEmail(fixture.ctx, user, "[email protected]")- return err- }- deleteInstance := func() error { return fixture.store.DeleteInstance(fixture.ctx, "default", "instance-a") }-- var firstResult, secondResult chan error- if first == "instance" {- firstResult, secondResult = make(chan error, 1), make(chan error, 1)- go func() { firstResult <- deleteInstance() }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- go func() { secondResult <- updateUser() }()- } else {- firstResult, secondResult = make(chan error, 1), make(chan error, 1)- go func() { firstResult <- updateUser() }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- go func() { secondResult <- deleteInstance() }()- }- waitForOperationBehindMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- barrier.release(t)-- requireMaintenanceResult(t, firstResult)- requireMaintenanceResult(t, secondResult)- requireUserInstanceDeletionState(fixture.ctx, t, fixture.db)- }-- t.Run("delete instance first", func(t *testing.T) { run(t, "instance") })- t.Run("update user first", func(t *testing.T) { run(t, "user") })-}diff --git a/backend/store/plan_check_run_lock_order_test.go b/backend/store/plan_check_run_lock_order_test.godeleted file mode 100644index a8c3d4a4b9..0000000000--- a/backend/store/plan_check_run_lock_order_test.go+++ /dev/null@@ -1,59 +0,0 @@-package store_test--import (- "context"- "testing"-- "github.com/stretchr/testify/require"-- "github.com/bytebase/bytebase/backend/common"- storepb "github.com/bytebase/bytebase/backend/generated-go/store"- "github.com/bytebase/bytebase/backend/store"-)--func TestCreatePlanCheckRunDoesNotDeadlockWithProjectDeletion(t *testing.T) {- err := runWithConcurrentProjectDeletion(t, `- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO plan_check_run (id, project, plan_id, status) VALUES (101, 'project-a', 101, 'FAILED');- `, "plan_check_run", 9902, func(ctx context.Context, s *store.Store) error {- _, err := s.CreatePlanCheckRun(ctx, &store.PlanCheckRunMessage{- ProjectID: "project-a",- PlanUID: 101,- Result: &storepb.PlanCheckRunResult{},- })- return err- })- require.NoError(t, err)-}--func TestCreateFirstPlanCheckRunStopsAfterProjectDeletion(t *testing.T) {- var created bool- err := runWithConcurrentProjectDeletion(t, `- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- `, "plan", 9904, func(ctx context.Context, s *store.Store) error {- var err error- created, err = s.CreatePlanCheckRun(ctx, &store.PlanCheckRunMessage{- ProjectID: "project-a",- PlanUID: 101,- Result: &storepb.PlanCheckRunResult{},- })- return err- })- require.NoError(t, err)- require.False(t, created)-}--func TestCreatePlanCheckRunRejectsDeletedProjectDuringProjectDeletion(t *testing.T) {- operationErr, deleteErr := runWithCreationBeforeProjectDeletion(t, `- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- `, "plan", func(ctx context.Context, s *store.Store) error {- _, err := s.CreatePlanCheckRun(ctx, &store.PlanCheckRunMessage{- ProjectID: "project-a",- PlanUID: 101,- Result: &storepb.PlanCheckRunResult{},- })- return err- })- require.NoError(t, deleteErr)- require.Equal(t, common.NotFound, common.ErrorCode(operationErr))-}diff --git a/backend/store/project.go b/backend/store/project.goindex a4e705e67a..58d590bbe7 100644--- a/backend/store/project.go+++ b/backend/store/project.go@@ -459,10 +459,6 @@ func (s *Store) DeleteProject(ctx context.Context, workspace string, resourceIDreturn errors.Wrap(err, "failed to begin transaction")}defer tx.Rollback()- if err := acquireProjectPurgeLock(ctx, tx, resourceID); err != nil {- return errors.Wrapf(err, "failed to lock project purge fence for %s", resourceID)- }-// Capture descendant cache keys before any descendant rows are removed so// the post-commit invalidation is precise and does not require re-reading// rows that no longer exist.@@ -471,8 +467,7 @@ func (s *Store) DeleteProject(ctx context.Context, workspace string, resourceIDreturn errors.Wrap(err, "failed to capture project descendant cache keys")}- // Delete query history before locking database-scoped rows to preserve the- // canonical sibling-branch order.+ // Delete query history before database-scoped rows.q := qb.Q().Space("DELETE FROM query_history WHERE project = ?", resourceID)sql, args, err := q.ToSQL()if err != nil {@@ -508,8 +503,6 @@ func (s *Store) DeleteProject(ctx context.Context, workspace string, resourceIDWHERE creator IN (SELECT email FROM service_account WHERE project = ? AND workspace = ?)OR creator IN (SELECT email FROM workload_identity WHERE project = ? AND workspace = ?))- ORDER BY saved_query, principal- FOR UPDATE)`, resourceID, workspace, resourceID, workspace, resourceID, workspace, resourceID, workspace)sql, args, err = q.ToSQL()if err != nil {@@ -611,40 +604,7 @@ func (s *Store) DeleteProject(ctx context.Context, workspace string, resourceIDreturn errors.Wrapf(err, "failed to delete task_run for project %s", resourceID)}- // Lock tasks in full primary-key order before deleting them. Pending Task Run- // creation uses the same order, so concurrent batches cannot form a wait cycle.- q = qb.Q().Space(`- SELECT project, id- FROM task- WHERE project = ?- ORDER BY project, id- FOR UPDATE`, resourceID)- sql, args, err = q.ToSQL()- if err != nil {- return errors.Wrap(err, "failed to build task lock query")- }- if err := func() error {- rows, err := tx.QueryContext(ctx, sql, args...)- if err != nil {- return errors.Wrapf(err, "failed to lock tasks for project %s", resourceID)- }- defer rows.Close()- for rows.Next() {- var lockedProjectID string- var lockedTaskID int64- if err := rows.Scan(&lockedProjectID, &lockedTaskID); err != nil {- return errors.Wrap(err, "failed to scan locked task")- }- }- if err := rows.Err(); err != nil {- return errors.Wrap(err, "failed to read locked tasks")- }- return nil- }(); err != nil {- return err- }-- // Delete tasks in plans of this project after acquiring every task lock.+ // Delete tasks in plans of this project.q = qb.Q().Space("DELETE FROM task WHERE project = ?", resourceID)sql, args, err = q.ToSQL()if err != nil {@@ -784,36 +744,6 @@ func (s *Store) DeleteProject(ctx context.Context, workspace string, resourceIDreturn errors.Wrapf(err, "failed to delete workload identities for project %s", resourceID)}- // Lock existing project instances after all of their descendants. A new- // project instance is rejected because this project is already soft-deleted.- q = qb.Q().Space(`- SELECT resource_id- FROM instance- WHERE project = ?- ORDER BY resource_id- FOR UPDATE- `, resourceID)- sql, args, err = q.ToSQL()- if err != nil {- return errors.Wrap(err, "failed to build project instance lock query")- }- if err := func() error {- rows, err := tx.QueryContext(ctx, sql, args...)- if err != nil {- return errors.Wrapf(err, "failed to lock project instances for project %s", resourceID)- }- defer rows.Close()- for rows.Next() {- var instanceID string- if err := rows.Scan(&instanceID); err != nil {- return errors.Wrap(err, "failed to scan locked project instance")- }- }- return rows.Err()- }(); err != nil {- return err- }-// Every project instance is deleted with its owner; they are never converted// into workspace instances.q = qb.Q().Space("DELETE FROM instance WHERE project = ?", resourceID)diff --git a/backend/store/query_history.go b/backend/store/query_history.goindex 364596ed3e..0b52807af7 100644--- a/backend/store/query_history.go+++ b/backend/store/query_history.go@@ -105,7 +105,7 @@ func (s *Store) CreateQueryHistory(ctx context.Context, create *QueryHistoryMessif err != nil {return nil, errors.Wrapf(err, "invalid query history database %q", create.Database)}- err = s.withDatabasePurgeFence(ctx, instanceID, databaseName, create.Project, nil, func(tx *sql.Tx, _ *databaseOwnership) error {+ err = s.withDatabaseWrite(ctx, instanceID, databaseName, nil, func(tx *sql.Tx, _ *databaseOwnership) error {return tx.QueryRowContext(ctx, query, args...).Scan(&create.ResourceID,&create.CreatedAt,diff --git a/backend/store/revision.go b/backend/store/revision.goindex 0fc4d302e8..374ae6eebb 100644--- a/backend/store/revision.go+++ b/backend/store/revision.go@@ -171,7 +171,7 @@ func (s *Store) CreateRevision(ctx context.Context, revision *RevisionMessage) (return nil, errors.Wrapf(err, "failed to marshal revision payload")}- err = s.withDatabasePurgeFence(ctx, revision.InstanceID, revision.DatabaseName, "", nil, func(tx *sql.Tx, _ *databaseOwnership) error {+ err = s.withDatabaseWrite(ctx, revision.InstanceID, revision.DatabaseName, nil, func(tx *sql.Tx, _ *databaseOwnership) error {return tx.QueryRowContext(ctx, query,revision.InstanceID,revision.DatabaseName,diff --git a/backend/store/sample_instance.go b/backend/store/sample_instance.goindex 8395fe9ebe..49798a8408 100644--- a/backend/store/sample_instance.go+++ b/backend/store/sample_instance.go@@ -183,14 +183,13 @@ func (s *Store) ActivateSampleInstanceSetup(ctx context.Context, workspaceID, refor _, projectID := range projects {var deleted boolif err := tx.QueryRowContext(ctx, `- SELECT deleted FROM project- WHERE workspace = $1 AND resource_id = $2- FOR UPDATE- `, workspaceID, projectID).Scan(&deleted); err != nil {+ SELECT deleted FROM project+ WHERE workspace = $1 AND resource_id = $2+ `, workspaceID, projectID).Scan(&deleted); err != nil {if errors.Is(err, sql.ErrNoRows) {return false, common.Errorf(common.NotFound, "project %s not found", projectID)}- return false, errors.Wrapf(err, "failed to lock project %s for sample instance setup activation", projectID)+ return false, errors.Wrapf(err, "failed to find project %s for sample instance setup activation", projectID)}if deleted {return false, common.Errorf(common.NotFound, "project %s is deleted", projectID)diff --git a/backend/store/sample_instance_test.go b/backend/store/sample_instance_test.goindex ee1158e2f6..6f5243e037 100644--- a/backend/store/sample_instance_test.go+++ b/backend/store/sample_instance_test.go@@ -2,20 +2,18 @@ package store_testimport ("context"- "database/sql""fmt""testing""time""github.com/stretchr/testify/require"- "github.com/bytebase/bytebase/backend/common""github.com/bytebase/bytebase/backend/common/testcontainer""github.com/bytebase/bytebase/backend/migrator""github.com/bytebase/bytebase/backend/store")-func newSampleInstanceFixture(t *testing.T) (context.Context, *sql.DB, *store.Store) {+func newSampleInstanceFixture(t *testing.T) (context.Context, *store.Store) {t.Helper()ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)t.Cleanup(cancel)@@ -40,11 +38,11 @@ func newSampleInstanceFixture(t *testing.T) (context.Context, *sql.DB, *store.Ststores, err := store.New(ctx, pgURL, false)require.NoError(t, err)t.Cleanup(func() { require.NoError(t, stores.Close()) })- return ctx, db, stores+ return ctx, stores}func TestSampleInstanceSetupPersistsOpaquePayload(t *testing.T) {- ctx, _, stores := newSampleInstanceFixture(t)+ ctx, stores := newSampleInstanceFixture(t)create := &store.SampleInstanceSetupMessage{WorkspaceID: "workspace-a",ReplicaID: "replica-a",@@ -75,7 +73,7 @@ func TestSampleInstanceSetupPersistsOpaquePayload(t *testing.T) {}func TestSampleInstanceSetupPermanentLifecycle(t *testing.T) {- ctx, _, stores := newSampleInstanceFixture(t)+ ctx, stores := newSampleInstanceFixture(t)setup := &store.SampleInstanceSetupMessage{WorkspaceID: "workspace-a",ReplicaID: "replica-a",@@ -111,7 +109,7 @@ func TestSampleInstanceSetupPermanentLifecycle(t *testing.T) {}func TestSampleInstanceSetupCleanupIsWorkspaceScoped(t *testing.T) {- ctx, _, stores := newSampleInstanceFixture(t)+ ctx, stores := newSampleInstanceFixture(t)now := time.Now().UTC().Truncate(time.Microsecond)for _, workspace := range []string{"workspace-a", "workspace-b"} {_, created, err := stores.ReserveSampleInstanceSetup(ctx, &store.SampleInstanceSetupMessage{@@ -155,7 +153,7 @@ func TestSampleInstanceSetupCleanupIsWorkspaceScoped(t *testing.T) {}func TestSampleInstanceSetupDeletedRowRemainsTombstone(t *testing.T) {- ctx, _, stores := newSampleInstanceFixture(t)+ ctx, stores := newSampleInstanceFixture(t)now := time.Now().UTC().Truncate(time.Microsecond)expiresAt := now.Add(time.Hour)setup := &store.SampleInstanceSetupMessage{@@ -198,153 +196,3 @@ func TestSampleInstanceSetupDeletedRowRemainsTombstone(t *testing.T) {require.False(t, created)require.Equal(t, deleted, again)}--func TestActivateSampleInstanceSetupSerializesWithProjectArchive(t *testing.T) {- ctx, db, stores := newSampleInstanceFixture(t)- _, created, err := stores.ReserveSampleInstanceSetup(ctx, &store.SampleInstanceSetupMessage{- WorkspaceID: "workspace-a",- ReplicaID: "replica-a",- Payload: []byte(`{"projectId":"project-workspace-a","instanceId":"sample-a"}`),- })- require.NoError(t, err)- require.True(t, created)-- archiveTx, err := db.BeginTx(ctx, nil)- require.NoError(t, err)- committed := false- defer func() {- if !committed {- _ = archiveTx.Rollback()- }- }()- _, err = archiveTx.ExecContext(ctx, `- UPDATE project SET deleted = TRUE- WHERE workspace = 'workspace-a' AND resource_id = 'project-workspace-a'- `)- require.NoError(t, err)-- result := make(chan error, 1)- go func() {- _, err := stores.ActivateSampleInstanceSetup(- ctx,- "workspace-a",- "replica-a",- []string{"project-workspace-a"},- time.Now(),- nil,- )- result <- err- }()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1 FROM pg_stat_activity- WHERE wait_event_type = 'Lock'- AND query LIKE '%SELECT deleted FROM project%FOR UPDATE%'- )- `).Scan(&waiting)- return err == nil && waiting- }, 5*time.Second, 10*time.Millisecond)- require.NoError(t, archiveTx.Commit())- committed = true- require.Equal(t, common.NotFound, common.ErrorCode(<-result))-- current, err := stores.GetSampleInstanceSetup(ctx, "workspace-a")- require.NoError(t, err)- require.Nil(t, current.ActivatedAt)-}--func TestProjectArchiveWaitsForSampleInstanceSetupActivation(t *testing.T) {- ctx, db, stores := newSampleInstanceFixture(t)- _, created, err := stores.ReserveSampleInstanceSetup(ctx, &store.SampleInstanceSetupMessage{- WorkspaceID: "workspace-a",- ReplicaID: "replica-a",- Payload: []byte(`{"projectId":"project-workspace-a","instanceId":"sample-a"}`),- })- require.NoError(t, err)- require.True(t, created)-- workspaceTx, err := db.BeginTx(ctx, nil)- require.NoError(t, err)- committed := false- defer func() {- if !committed {- _ = workspaceTx.Rollback()- }- }()- var lockedWorkspace string- err = workspaceTx.QueryRowContext(ctx, `- SELECT resource_id FROM workspace- WHERE resource_id = 'workspace-a'- FOR UPDATE- `).Scan(&lockedWorkspace)- require.NoError(t, err)-- activationResult := make(chan struct {- activated bool- err error- }, 1)- go func() {- activated, err := stores.ActivateSampleInstanceSetup(- ctx,- "workspace-a",- "replica-a",- []string{"project-workspace-a"},- time.Now(),- nil,- )- activationResult <- struct {- activated bool- err error- }{activated: activated, err: err}- }()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1 FROM pg_stat_activity- WHERE wait_event_type = 'Lock'- AND query LIKE '%SELECT deleted FROM workspace%FOR UPDATE%'- )- `).Scan(&waiting)- return err == nil && waiting- }, 5*time.Second, 10*time.Millisecond)-- archiveResult := make(chan error, 1)- go func() {- _, err := db.ExecContext(ctx, `- UPDATE project SET deleted = TRUE- WHERE workspace = 'workspace-a' AND resource_id = 'project-workspace-a'- `)- archiveResult <- err- }()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1 FROM pg_stat_activity- WHERE wait_event_type = 'Lock'- AND query LIKE '%UPDATE project SET deleted = TRUE%'- )- `).Scan(&waiting)- return err == nil && waiting- }, 5*time.Second, 10*time.Millisecond)-- require.NoError(t, workspaceTx.Commit())- committed = true- activation := <-activationResult- require.NoError(t, activation.err)- require.True(t, activation.activated)- require.NoError(t, <-archiveResult)-- current, err := stores.GetSampleInstanceSetup(ctx, "workspace-a")- require.NoError(t, err)- require.NotNil(t, current.ActivatedAt)- var projectDeleted bool- require.NoError(t, db.QueryRowContext(ctx, `- SELECT deleted FROM project- WHERE workspace = 'workspace-a' AND resource_id = 'project-workspace-a'- `).Scan(&projectDeleted))- require.True(t, projectDeleted)-}diff --git a/backend/store/saved_query.go b/backend/store/saved_query.goindex 82ba7cc3fa..78c93ff5ee 100644--- a/backend/store/saved_query.go+++ b/backend/store/saved_query.go@@ -378,10 +378,7 @@ func (s *Store) SetSavedQueryBindings(ctx context.Context, projectID, resourceID// compare-and-swap write; the caller refetches and reapplies.var ErrSavedQueryEtagMismatch = errors.New("saved query policy etag mismatch")-// CreateSavedQuery creates a new saved query. The insert transaction locks-// the owning project row and requires it to be active, so a create racing a-// project purge fails cleanly instead of as an FK violation — the parent-// fence for a new child row that cannot be locked in advance.+// CreateSavedQuery creates a new saved query under an active project.func (s *Store) CreateSavedQuery(ctx context.Context, create *SavedQueryMessage) (*SavedQueryMessage, error) {payloadStr, err := protojson.Marshal(&storepb.SavedQueryPayload{Database: create.Database})if err != nil {@@ -394,9 +391,8 @@ func (s *Store) CreateSavedQuery(ctx context.Context, create *SavedQueryMessage)}defer tx.Rollback()- // A saved query is a new child row, so it takes the "requires an active- // project" fence: lock the parent and refuse if it is archived or purged.- if err := lockActiveProject(ctx, tx, create.ProjectID); err != nil {+ // A saved query is a new child row, so its project must be active.+ if err := requireActiveProject(ctx, tx, create.ProjectID); err != nil {return nil, err}@@ -470,24 +466,15 @@ func (s *Store) PatchSavedQuery(ctx context.Context, patch *PatchSavedQueryMessa// DeleteSavedQuery deletes an existing saved query and reports whether it was// still there to delete. Star rows are deleted first, in full primary-key// order — explicitly, not via the FK cascade (which would lock the parent-// first) — matching the star and purge lock order so a delete racing a star-// toggle or a purge cannot deadlock.+// first) — so a delete racing a star toggle cannot deadlock.//-// Both statements scope by the project the caller was authorized in: a purge-// reassigns surviving saved queries to the default project, and a delete-// racing it must not land on the reassigned row. The row delete is the-// arbiter — when it matches nothing (row gone, or reassigned after the star-// statement's unlocked parent snapshot), the transaction rolls back, which-// also restores any stars the first statement removed, and the caller gets-// false to answer NotFound.+// Both statements scope by the project the caller was authorized in. The row+// delete is the arbiter — when it matches nothing, the transaction rolls back,+// which also restores any stars the first statement removed, and the caller+// gets false to answer NotFound.//-// Lifecycle policy (per the store's purge-fence rule): writers on existing-// rows — delete, patch, star — require the row in the authorized project,-// not an active project. Archived projects are unreachable through every-// read path (the project.deleted = FALSE fence on the fetches), so the only-// archival exposure is a write already in flight when the archive lands,-// and that completing is an ordinary serialization of concurrent requests.-// Only creation, which adds a row a purge cannot see, requires an active+// Writers on existing rows — delete, patch, star — require the row in the+// authorized project, not an active project. Only creation requires an active// project.func (s *Store) DeleteSavedQuery(ctx context.Context, projectID, resourceID string) (bool, error) {tx, err := s.GetDB().BeginTx(ctx, nil)@@ -527,17 +514,12 @@ func (s *Store) DeleteSavedQuery(ctx context.Context, projectID, resourceID stri}// SetSavedQueryStar stars or unstars a saved query for a principal, and-// reports whether the saved query was still there to star. Lock order per the-// store row-lock rules: toggling or removing an existing star locks that child-// row directly; only the first star for a (query, principal) inserts a child-// that cannot be locked in advance — that case alone takes the parent fence-// (lock the saved_query row), and its inserted key is novel so it never-// contends with a purge's existing-child locks. The parent fence is also the-// only path that can observe a concurrent delete or purge reassignment — it-// requires the row in the project the caller was authorized in — and reports-// either as false rather than an error so the caller decides what a vanished-// row means. Removing an existing star stays project-blind: the star is the-// caller's own marker, removable wherever its row went.+// reports whether the saved query was still there to star. Toggling or removing+// an existing star locks that row directly. The first star instead locks the+// saved query row before inserting, so it serializes with deletion and verifies+// that the query still belongs to the project the caller was authorized in.+// Removing an existing star stays project-blind: the star is the caller's own+// marker, removable wherever its row went.func (s *Store) SetSavedQueryStar(ctx context.Context, projectID, savedQueryResourceID, principal string, starred bool) (bool, error) {tx, err := s.GetDB().BeginTx(ctx, nil)if err != nil {diff --git a/backend/store/saved_query_bindings_test.go b/backend/store/saved_query_bindings_test.goindex af3e098c7e..9f69dceb53 100644--- a/backend/store/saved_query_bindings_test.go+++ b/backend/store/saved_query_bindings_test.go@@ -19,7 +19,7 @@ func TestSavedQueryBindingsStoredShape(t *testing.T) {INSERT INTO saved_query (resource_id, creator, project, name, statement)VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', 'SELECT 1;');`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)bindings := []*storepb.SavedQueryBinding{{Level: storepb.SavedQueryBinding_EDITOR,@@ -102,7 +102,7 @@ func TestSavedQueryBindingsFollowGroupRename(t *testing.T) {INSERT INTO saved_query (resource_id, creator, project, name, statement)VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', 'SELECT 1;');`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)emptyEtag, err := store.SavedQueryPolicyEtag(nil)require.NoError(t, err)diff --git a/backend/store/saved_query_lock_order_test.go b/backend/store/saved_query_lock_order_test.goindex 9d1e5a842f..7a762b1c22 100644--- a/backend/store/saved_query_lock_order_test.go+++ b/backend/store/saved_query_lock_order_test.go@@ -7,17 +7,10 @@ import ("github.com/stretchr/testify/require""github.com/bytebase/bytebase/backend/common"- storepb "github.com/bytebase/bytebase/backend/generated-go/store""github.com/bytebase/bytebase/backend/store")-// Deterministic regression locks for the saved-query lock-order invariants:-// child-before-parent for existing rows, parent fence only for a new child-// insert, and full primary-key order for every multi-row lock. Each pair is-// exercised in both acquisition directions, asserting terminal outcomes —-// never a deadlock (40P01) or an FK failure.--func savedQueryStarCount(t *testing.T, fixture *projectDeletionLockOrderFixture) int {+func savedQueryStarCount(t *testing.T, fixture *storePostgresFixture) int {t.Helper()var count intrequire.NoError(t, fixture.db.QueryRowContext(fixture.ctx,@@ -25,144 +18,6 @@ func savedQueryStarCount(t *testing.T, fixture *projectDeletionLockOrderFixture)return count}-func TestSavedQueryCreateAndDeleteProjectLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO saved_query (resource_id, creator, project, name, statement)- VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '');- `-- t.Run("create rejects a deleted project while its purge is parked", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9931- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER UPDATE OF project ON saved_query FOR EACH ROW")-- deleteResult := make(chan error, 1)- go func() { deleteResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- // The create fence locks the project row (which the parked purge does- // not yet hold — the purge locks parents last), sees the deleted flag,- // and rejects cleanly instead of racing to an FK failure.- createResult := make(chan error, 1)- go func() {- _, err := fixture.store.CreateSavedQuery(fixture.ctx, &store.SavedQueryMessage{- ProjectID: "project-a",- Creator: "[email protected]",- Title: "raced",- Statement: "SELECT 1;",- })- createResult <- err- }()- select {- case err := <-createResult:- require.Error(t, err)- require.ErrorContains(t, err, "deleted")- case <-time.After(maintenanceLockWait):- t.Fatal("saved query creation should reject a deleted project before purge completion")- }- barrier.release(t)- requireMaintenanceResult(t, deleteResult)- })-- t.Run("create succeeds on an unrelated project while a purge is parked", func(t *testing.T) {- // project-b is untouched by project-a's purge. The default project is- // deliberately not the target here: the purge re-parents saved queries- // into it, and that FK write holds a KEY SHARE lock on the default- // project row for the rest of the purge transaction, so a create fence- // there legitimately waits for the (short-lived) purge to commit.- fixture := newProjectDeletionLockOrderFixture(t, seedSQL+`- INSERT INTO project (resource_id, workspace, name) VALUES ('project-b', 'default', 'Project B');- `)- const barrierID = 9932- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER UPDATE OF project ON saved_query FOR EACH ROW")-- deleteResult := make(chan error, 1)- go func() { deleteResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)-- created, err := fixture.store.CreateSavedQuery(fixture.ctx, &store.SavedQueryMessage{- ProjectID: "project-b",- Creator: "[email protected]",- Title: "independent",- Statement: "SELECT 1;",- })- require.NoError(t, err)- require.NotEmpty(t, created.ResourceID)-- barrier.release(t)- requireMaintenanceResult(t, deleteResult)- })-}--func TestSavedQueryFirstStarAndDeleteProjectLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO saved_query (resource_id, creator, project, name, statement)- VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '');- `-- t.Run("purge first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9933- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER UPDATE OF project ON saved_query FOR EACH ROW")-- deleteResult := make(chan error, 1)- go func() { deleteResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- purgePID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- // The first star takes the parent fence on the saved_query row the- // parked purge holds via its re-parent update.- starApplied := make(chan bool, 1)- starResult := make(chan error, 1)- go func() {- applied, err := fixture.store.SetSavedQueryStar(fixture.ctx, "project-a", "saved-query-a", "[email protected]", true)- starApplied <- applied- starResult <- err- }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, purgePID)- barrier.release(t)-- requireMaintenanceResult(t, deleteResult)- requireMaintenanceResult(t, starResult)- // The row survived the purge by re-parenting into the default project,- // so the fence re-evaluates against the committed reassignment and the- // star authorized in project-a does not land.- require.False(t, <-starApplied)- require.Equal(t, 0, savedQueryStarCount(t, fixture))- })-- t.Run("star first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9934- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER INSERT ON saved_query_star FOR EACH ROW")-- starResult := make(chan error, 1)- go func() {- _, err := fixture.store.SetSavedQueryStar(fixture.ctx, "project-a", "saved-query-a", "[email protected]", true)- starResult <- err- }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- starPID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- deleteResult := make(chan error, 1)- go func() { deleteResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, starPID)- barrier.release(t)-- requireMaintenanceResult(t, starResult)- requireMaintenanceResult(t, deleteResult)- require.Equal(t, 1, savedQueryStarCount(t, fixture))- })-}-func TestSavedQueryDeleteAndStarToggleLockOrder(t *testing.T) {const seedSQL = `INSERT INTO saved_query (resource_id, creator, project, name, statement)@@ -171,12 +26,9 @@ func TestSavedQueryDeleteAndStarToggleLockOrder(t *testing.T) {`t.Run("delete first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)const barrierID = 9935barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- // Park the delete after it has removed the star rows and taken the- // parent row itself — that is the point where a first star from another- // principal must be fenced out.installMaintenanceLockBarrier(t, fixture.db, barrierID,"AFTER DELETE ON saved_query FOR EACH ROW")@@ -188,8 +40,6 @@ func TestSavedQueryDeleteAndStarToggleLockOrder(t *testing.T) {waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)deletePID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)- // A second principal's first star waits on the parent fence held by- // the parked delete, then reports the parent gone once it lands.starResult := make(chan error, 1)var starApplied boolgo func() {@@ -203,8 +53,6 @@ func TestSavedQueryDeleteAndStarToggleLockOrder(t *testing.T) {requireMaintenanceResult(t, deleteResult)select {case err := <-starResult:- // The parent is gone, so there is nothing to star: reported as- // not applied rather than as a failure.require.NoError(t, err)require.False(t, starApplied)case <-time.After(maintenanceLockWait):@@ -214,14 +62,12 @@ func TestSavedQueryDeleteAndStarToggleLockOrder(t *testing.T) {})t.Run("toggle first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)const barrierID = 9936barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)installMaintenanceLockBarrier(t, fixture.db, barrierID,"AFTER DELETE ON saved_query_star FOR EACH ROW")- // Unstar locks and deletes the existing child row, parking at the- // barrier with the child lock held.unstarResult := make(chan error, 1)go func() {_, err := fixture.store.SetSavedQueryStar(fixture.ctx, "default", "saved-query-a", "[email protected]", false)@@ -247,80 +93,6 @@ func TestSavedQueryDeleteAndStarToggleLockOrder(t *testing.T) {})}-func TestSavedQueryDeleteAndDeleteProjectLockOrder(t *testing.T) {- // The saved query is owned by a project service account, so the purge- // deletes it (stars first) rather than re-parenting it — the contended- // pair is the star child rows, locked in primary-key order on both paths.- const seedSQL = `- INSERT INTO service_account (name, email, workspace, service_key_hash, project)- VALUES ('service account', '[email protected]', 'default', 'unused', 'project-a');- INSERT INTO saved_query (resource_id, creator, project, name, statement)- VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '');- INSERT INTO saved_query_star (saved_query, principal) VALUES- ('saved-query-a', '[email protected]'),- ('saved-query-a', '[email protected]');- `-- requireGone := func(t *testing.T, fixture *projectDeletionLockOrderFixture) {- t.Helper()- var queryCount int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM saved_query WHERE resource_id = 'saved-query-a'").Scan(&queryCount))- require.Zero(t, queryCount)- require.Zero(t, savedQueryStarCount(t, fixture))- }-- t.Run("purge first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9937- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER DELETE ON saved_query_star FOR EACH ROW")-- purgeResult := make(chan error, 1)- go func() { purgeResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- purgePID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- deleteResult := make(chan error, 1)- go func() {- _, err := fixture.store.DeleteSavedQuery(fixture.ctx, "project-a", "saved-query-a")- deleteResult <- err- }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, purgePID)- barrier.release(t)-- requireMaintenanceResult(t, purgeResult)- requireMaintenanceResult(t, deleteResult)- requireGone(t, fixture)- })-- t.Run("delete first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9938- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER DELETE ON saved_query_star FOR EACH ROW")-- deleteResult := make(chan error, 1)- go func() {- _, err := fixture.store.DeleteSavedQuery(fixture.ctx, "project-a", "saved-query-a")- deleteResult <- err- }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- deletePID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- purgeResult := make(chan error, 1)- go func() { purgeResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, deletePID)- barrier.release(t)-- requireMaintenanceResult(t, deleteResult)- requireMaintenanceResult(t, purgeResult)- requireGone(t, fixture)- })-}-func TestSavedQueryBatchFolderMovesLockOrder(t *testing.T) {const seedSQL = `INSERT INTO saved_query (resource_id, creator, project, name, statement, folder) VALUES@@ -328,20 +100,14 @@ func TestSavedQueryBatchFolderMovesLockOrder(t *testing.T) {('saved-query-b', '[email protected]', 'default', 'B', '', 'start/sub');`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)const barrierID = 9939barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER UPDATE OF folder ON saved_query FOR EACH ROW")+ installMaintenanceLockBarrier(t, fixture.db, barrierID, "AFTER UPDATE OF folder ON saved_query FOR EACH ROW")- // Two overlapping folder moves over the same rows: both lock in full- // primary-key order, so the second serializes behind the first instead of- // deadlocking, re-evaluates its predicate against the committed rename,- // and moves nothing.firstResult := make(chan error, 1)go func() {- _, err := fixture.store.MoveSavedQueryFolder(fixture.ctx, "default", "[email protected]",- "start", "one")+ _, err := fixture.store.MoveSavedQueryFolder(fixture.ctx, "default", "[email protected]", "start", "one")firstResult <- err}()waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)@@ -350,14 +116,12 @@ func TestSavedQueryBatchFolderMovesLockOrder(t *testing.T) {secondMoved := make(chan int, 1)secondResult := make(chan error, 1)go func() {- moved, err := fixture.store.MoveSavedQueryFolder(fixture.ctx, "default", "[email protected]",- "start", "two")+ moved, err := fixture.store.MoveSavedQueryFolder(fixture.ctx, "default", "[email protected]", "start", "two")secondMoved <- movedsecondResult <- err}()waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, firstPID)barrier.release(t)-requireMaintenanceResult(t, firstResult)requireMaintenanceResult(t, secondResult)require.Equal(t, 0, <-secondMoved)@@ -376,14 +140,9 @@ func TestSavedQueryBatchFolderMovesLockOrder(t *testing.T) {require.Equal(t, []string{"one", "one/sub"}, folders)}-// TestCreateSavedQueryRejectsInactiveProject covers the shared fence's outcome-// on this path: the API layer checks the parent before calling, so the fence-// only fires when the project is archived or purged in between, and the caller-// has to be able to tell that from a genuine failure.func TestCreateSavedQueryRejectsInactiveProject(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, "")+ fixture := newStorePostgresFixture(t, "")- // project-a is seeded soft-deleted._, err := fixture.store.CreateSavedQuery(fixture.ctx, &store.SavedQueryMessage{ProjectID: "project-a",Creator: "[email protected]",@@ -402,7 +161,6 @@ func TestCreateSavedQueryRejectsInactiveProject(t *testing.T) {require.Error(t, err)require.Equal(t, common.NotFound, common.ErrorCode(err))- // An active parent is unaffected.created, err := fixture.store.CreateSavedQuery(fixture.ctx, &store.SavedQueryMessage{ProjectID: "default",Creator: "[email protected]",@@ -413,100 +171,6 @@ func TestCreateSavedQueryRejectsInactiveProject(t *testing.T) {require.NotEmpty(t, created.ResourceID)}-// SetSavedQueryBindings updates one existing row and touches no child table, so-// it cannot form a wait-for cycle with the purge by itself. What it can do is-// race project deletion, and both orders must reach a terminal outcome: either-// the policy lands and the purge then removes the row, or the purge wins and the-// write reports the saved query as gone rather than failing on the FK.-func TestSavedQuerySetBindingsAndDeleteProjectLockOrder(t *testing.T) {- const seedSQL = `- INSERT INTO service_account (name, email, workspace, service_key_hash, project)- VALUES ('service account', '[email protected]', 'default', 'unused', 'project-a');- INSERT INTO saved_query (resource_id, creator, project, name, statement)- VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '');- INSERT INTO saved_query_star (saved_query, principal) VALUES- ('saved-query-a', '[email protected]');- `-- bindings := []*storepb.SavedQueryBinding{{- Level: storepb.SavedQueryBinding_VIEWER,- Members: []string{"user:[email protected]"},- }}-- requireGone := func(t *testing.T, fixture *projectDeletionLockOrderFixture) {- t.Helper()- var count int- require.NoError(t, fixture.db.QueryRowContext(fixture.ctx,- "SELECT COUNT(*) FROM saved_query WHERE resource_id = 'saved-query-a'").Scan(&count))- require.Zero(t, count, "the purge owns the terminal state in both orders")- require.Zero(t, savedQueryStarCount(t, fixture))- }-- emptyEtag, err := store.SavedQueryPolicyEtag(nil)- require.NoError(t, err)-- t.Run("purge first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9951- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- // The barrier has to fire where the purge actually holds the- // saved_query row: pausing it after the star delete leaves that row- // unlocked, so the policy write would sail past and prove nothing.- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER DELETE ON saved_query FOR EACH ROW")-- purgeResult := make(chan error, 1)- go func() { purgeResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- purgePID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- setResult := make(chan error, 1)- go func() {- _, err := fixture.store.SetSavedQueryBindings(fixture.ctx, "project-a", "saved-query-a", bindings, emptyEtag)- setResult <- err- }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, purgePID)- barrier.release(t)-- requireMaintenanceResult(t, purgeResult)- // The row is gone by the time the lock is granted, which the store- // reports as "not applied" (a nil error) rather than an FK failure.- requireMaintenanceResult(t, setResult)- requireGone(t, fixture)- })-- t.Run("policy write first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)- const barrierID = 9952- barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)- installMaintenanceLockBarrier(t, fixture.db, barrierID,- "AFTER UPDATE OF bindings ON saved_query FOR EACH ROW")-- setResult := make(chan error, 1)- go func() {- _, err := fixture.store.SetSavedQueryBindings(fixture.ctx, "project-a", "saved-query-a", bindings, emptyEtag)- setResult <- err- }()- waitForMaintenanceBarrier(fixture.ctx, t, fixture.db, barrierID)- setPID := maintenanceBarrierWaitingPID(fixture.ctx, t, fixture.db, barrierID)-- purgeResult := make(chan error, 1)- go func() { purgeResult <- fixture.store.DeleteProject(fixture.ctx, "default", "project-a") }()- waitForBackendBlockedByPID(fixture.ctx, t, fixture.db, setPID)- barrier.release(t)-- requireMaintenanceResult(t, setResult)- requireMaintenanceResult(t, purgeResult)- requireGone(t, fixture)- })-}--// A prefix move locks every row under the folder, so two moves whose paths-// overlap contend on the same rows -- "a" covers "a/child". Both take their-// locks in primary-key order through the CTE, so the second serializes behind-// the first in either acquisition order rather than deadlocking, and the-// terminal folders are exact: whichever commits first decides what the other-// still matches.func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {const seedSQL = `INSERT INTO saved_query (resource_id, creator, project, name, statement, folder) VALUES@@ -514,7 +178,7 @@ func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {('saved-query-b', '[email protected]', 'default', 'B', '', 'a/child');`- folders := func(t *testing.T, fixture *projectDeletionLockOrderFixture) map[string]string {+ folders := func(t *testing.T, fixture *storePostgresFixture) map[string]string {t.Helper()rows, err := fixture.db.QueryContext(fixture.ctx,"SELECT resource_id, folder FROM saved_query ORDER BY resource_id")@@ -531,7 +195,7 @@ func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {}t.Run("parent first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)const barrierID = 9961barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)installMaintenanceLockBarrier(t, fixture.db, barrierID,@@ -555,8 +219,6 @@ func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {requireMaintenanceResult(t, parentResult)requireMaintenanceResult(t, childResult)- // The parent move carried the child with it, so "a/child" no longer- // exists for the second move to find.require.Equal(t, map[string]string{"saved-query-a": "x","saved-query-b": "x/child",@@ -564,7 +226,7 @@ func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {})t.Run("child first", func(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)const barrierID = 9962barrier := newMaintenanceLockBarrier(fixture.ctx, t, fixture.db, barrierID)installMaintenanceLockBarrier(t, fixture.db, barrierID,@@ -588,7 +250,6 @@ func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {requireMaintenanceResult(t, childResult)requireMaintenanceResult(t, parentResult)- // The child left the subtree first, so the parent move only found "a".require.Equal(t, map[string]string{"saved-query-a": "x","saved-query-b": "y",@@ -596,20 +257,14 @@ func TestSavedQueryFolderPrefixMovesLockOrder(t *testing.T) {})}-// TestSavedQueryWritersProjectScoped locks the purge-reassignment fence: a-// purge moves surviving saved queries to the default project, so a write-// authorized in the original project must not land on the reassigned row.-// Every non-purge writer scopes by (resource_id, project).func TestSavedQueryWritersProjectScoped(t *testing.T) {const seedSQL = `INSERT INTO saved_query (resource_id, creator, project, name, statement, folder)VALUES ('saved-query-a', '[email protected]', 'project-a', 'Saved Query A', '', 'kept');INSERT INTO saved_query_star (saved_query, principal) VALUES ('saved-query-a', '[email protected]');`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)- // Simulate the purge's re-parenting having happened after the caller's- // authorization in project-a._, err := fixture.db.ExecContext(fixture.ctx,"UPDATE saved_query SET project = 'default' WHERE resource_id = 'saved-query-a'")require.NoError(t, err)diff --git a/backend/store/saved_query_purge_isolation_test.go b/backend/store/saved_query_purge_isolation_test.goindex a45ef6fc6a..d7d12a7c59 100644--- a/backend/store/saved_query_purge_isolation_test.go+++ b/backend/store/saved_query_purge_isolation_test.go@@ -32,7 +32,7 @@ func TestDeleteProjectPurgeKeepsOtherProjectSavedQueryStars(t *testing.T) {('sq-b-user', '[email protected]');`- fixture := newProjectDeletionLockOrderFixture(t, seedSQL)+ fixture := newStorePostgresFixture(t, seedSQL)require.NoError(t, fixture.store.DeleteProject(fixture.ctx, "default", "project-a"))savedQueryExists := func(resourceID string) bool {diff --git a/backend/store/sheet.go b/backend/store/sheet.goindex 4c8ea4a15e..fdd4ba510b 100644--- a/backend/store/sheet.go+++ b/backend/store/sheet.go@@ -256,12 +256,7 @@ func (s *Store) filterSheetsForProject(ctx context.Context, projectID string, sh// Duplicate statements share the same blob (ON CONFLICT DO NOTHING); refs are// per project.//-// Lifecycle policy: creating a sheet requires an active project. The-// transaction takes the project purge fence before any row lock, then locks-// the project row and rejects a missing or deleted project with NotFound,-// serializing against project purge in both directions. Both inserts are-// new-row-only; the foreign-key checks on project and sheet_blob are covered-// by the fence and the project row lock.+// Creating a sheet requires an active project. Both inserts are new-row-only.func (s *Store) CreateSheets(ctx context.Context, projectID string, creates ...*SheetMessage) ([]*SheetMessage, error) {if projectID == "" {return nil, errors.New("project is required to create sheets")@@ -286,10 +281,7 @@ func (s *Store) CreateSheets(ctx context.Context, projectID string, creates ...*}defer tx.Rollback()- if err := acquireProjectPurgeLock(ctx, tx, projectID); err != nil {- return nil, errors.Wrapf(err, "failed to lock project purge fence for %s", projectID)- }- if err := lockActiveProject(ctx, tx, projectID); err != nil {+ if err := requireActiveProject(ctx, tx, projectID); err != nil {return nil, err}diff --git a/backend/store/sheet_lock_order_test.go b/backend/store/sheet_lock_order_test.godeleted file mode 100644index 23dd82c3c8..0000000000--- a/backend/store/sheet_lock_order_test.go+++ /dev/null@@ -1,44 +0,0 @@-package store_test--import (- "context"- "strings"- "testing"-- "github.com/stretchr/testify/require"-- "github.com/bytebase/bytebase/backend/common"- "github.com/bytebase/bytebase/backend/store"-)--func createSheetInProjectA(ctx context.Context, s *store.Store) error {- _, err := s.CreateSheets(ctx, "project-a", &store.SheetMessage{Statement: "SELECT 1;"})- return err-}--const sheetLockOrderSeedSQL = `- INSERT INTO sheet_blob (sha256, content) VALUES (sha256('SELECT seed;'::bytea), 'SELECT seed;');- INSERT INTO sheet_blob_ref (project, sha256) VALUES ('project-a', sha256('SELECT seed;'::bytea));-`--// Purge first: project deletion wedges inside its sheet_blob_ref delete while-// CreateSheets queues behind the project purge fence. The create must end in-// a clean NotFound once the purge wins — no deadlock, no FK failure — and the-// purge itself must succeed against the seeded refs.-func TestCreateSheetsDoesNotDeadlockWithProjectDeletion(t *testing.T) {- err := runWithConcurrentProjectDeletion(t, sheetLockOrderSeedSQL, "sheet_blob_ref", 9905, createSheetInProjectA)- require.Equal(t, common.NotFound, common.ErrorCode(err))- require.NotContains(t, strings.ToLower(err.Error()), "foreign key",- "the create must reject the purged project cleanly, not with an FK failure")-}--// Writer first: CreateSheets holds the purge fence and waits on the project-// row; project deletion queues behind the fence. The create must reject the-// soft-deleted project with NotFound, and the purge must then complete.-func TestCreateSheetsRejectsDeletedProjectDuringProjectDeletion(t *testing.T) {- operationErr, deleteErr := runWithCreationBeforeProjectDeletion(t, sheetLockOrderSeedSQL, "sheet_blob_ref", createSheetInProjectA)- require.NoError(t, deleteErr)- require.Equal(t, common.NotFound, common.ErrorCode(operationErr))- require.NotContains(t, strings.ToLower(operationErr.Error()), "foreign key",- "the create must reject the deleted project cleanly, not with an FK failure")-}diff --git a/backend/store/sync_history.go b/backend/store/sync_history.goindex 0e33c56e63..ae72c7ae03 100644--- a/backend/store/sync_history.go+++ b/backend/store/sync_history.go@@ -94,7 +94,7 @@ func (s *Store) CreateSyncHistory(ctx context.Context, instanceID, databaseName}var resourceID string- err = s.withDatabasePurgeFence(ctx, instanceID, databaseName, "", nil, func(tx *sql.Tx, _ *databaseOwnership) error {+ err = s.withDatabaseWrite(ctx, instanceID, databaseName, nil, func(tx *sql.Tx, _ *databaseOwnership) error {return tx.QueryRowContext(ctx, query, args...).Scan(&resourceID)})if err != nil {diff --git a/backend/store/task_run.go b/backend/store/task_run.goindex dbab7f0fb3..0f05ff8268 100644--- a/backend/store/task_run.go+++ b/backend/store/task_run.go@@ -355,12 +355,6 @@ func (s *Store) CreatePendingTaskRuns(ctx context.Context, creator string, creatreturn errors.Wrapf(err, "failed to begin tx")}defer tx.Rollback()- for _, instanceID := range instanceIDs {- if err := acquireInstancePurgeLock(ctx, tx, instanceID); err != nil {- return errors.Wrapf(err, "failed to lock instance lifecycle fence for %s", instanceID)- }- }-lockQ := qb.Q().Space(`SELECT task.project, task.id, task.instanceFROM (@@ -411,19 +405,19 @@ func (s *Store) CreatePendingTaskRuns(ctx context.Context, creator string, creatfor _, instanceID := range lockedInstanceIDs {var deleted boolif err := tx.QueryRowContext(ctx, `- SELECT deleted FROM instance WHERE resource_id = $1 FOR UPDATE+ SELECT deleted FROM instance WHERE resource_id = $1`, instanceID).Scan(&deleted); err != nil {if errors.Is(err, sql.ErrNoRows) {return common.Errorf(common.NotFound, "instance %s not found", instanceID)}- return errors.Wrapf(err, "failed to lock instance %s", instanceID)+ return errors.Wrapf(err, "failed to find instance %s", instanceID)}if deleted {return common.Errorf(common.Conflict, "instance %s is archived", instanceID)}}- // Keep the child-to-parent lock order used by project deletion: task, instance, then project.+ // nextProjectID locks the project row to serialize ID allocation.baseID, err := nextProjectID(ctx, tx, "task_run", projectID)if err != nil {return errdiff --git a/backend/store/task_run_lock_order_test.go b/backend/store/task_run_claim_test.gosimilarity index 52%rename from backend/store/task_run_lock_order_test.gorename to backend/store/task_run_claim_test.goindex 1ddac9f5d7..e4f705e6a1 100644--- a/backend/store/task_run_lock_order_test.go+++ b/backend/store/task_run_claim_test.go@@ -9,14 +9,13 @@ import ("github.com/stretchr/testify/require"- "github.com/bytebase/bytebase/backend/common""github.com/bytebase/bytebase/backend/common/testcontainer""github.com/bytebase/bytebase/backend/migrator""github.com/bytebase/bytebase/backend/store")func TestClaimAvailableTaskRunsSkipsDeletedProjects(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `+ fixture := newStorePostgresFixture(t, `INSERT INTO instance (resource_id, workspace) VALUES ('instance-a', 'default');INSERT INTO plan (id, creator, project, name, description)VALUES (101, '[email protected]', 'project-a', 'Plan A', '');@@ -44,7 +43,7 @@ func TestClaimAvailableTaskRunsSkipsDeletedProjects(t *testing.T) {}func TestClaimAvailableTaskRunsSkipsArchivedInstances(t *testing.T) {- fixture := newTaskRunArchivalLockOrderFixture(t, `+ fixture := newTaskRunClaimFixture(t, `INSERT INTO instance (resource_id, workspace, deleted) VALUES ('instance-a', 'default', TRUE);INSERT INTO plan (id, creator, project, name, description)VALUES (101, '[email protected]', 'project-a', 'Plan A', '');@@ -72,7 +71,7 @@ func TestClaimAvailableTaskRunsSkipsArchivedInstances(t *testing.T) {}func TestClaimAvailableTaskRunsClaimsLiveInstanceSpecialTaskTypes(t *testing.T) {- fixture := newTaskRunArchivalLockOrderFixture(t, `+ fixture := newTaskRunClaimFixture(t, `INSERT INTO instance (resource_id, workspace) VALUES ('instance-a', 'default');INSERT INTO plan (id, creator, project, name, description)VALUES (101, '[email protected]', 'project-a', 'Plan A', '');@@ -97,7 +96,7 @@ func TestClaimAvailableTaskRunsClaimsLiveInstanceSpecialTaskTypes(t *testing.T)require.Equal(t, "RUNNING", status)}-func newTaskRunArchivalLockOrderFixture(t *testing.T, seedSQL string) *projectDeletionLockOrderFixture {+func newTaskRunClaimFixture(t *testing.T, seedSQL string) *storePostgresFixture {t.Helper()ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)t.Cleanup(cancel)@@ -121,110 +120,5 @@ func newTaskRunArchivalLockOrderFixture(t *testing.T, seedSQL string) *projectDerequire.NoError(t, err)t.Cleanup(func() { require.NoError(t, s.Close()) })- return &projectDeletionLockOrderFixture{ctx: ctx, db: db, store: s}-}--func TestCreatePendingTaskRunsDoesNotDeadlockWithProjectDeletion(t *testing.T) {- fixture := newProjectDeletionLockOrderFixture(t, `- INSERT INTO instance (resource_id, workspace) VALUES ('instance-a', 'default');- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO task (id, project, plan_id, instance, type) VALUES (102, 'project-a', 101, 'instance-a', 'DATABASE_SCHEMA_UPDATE');- INSERT INTO task (id, project, plan_id, instance, type) VALUES (101, 'project-a', 101, 'instance-a', 'DATABASE_SCHEMA_UPDATE');- `)- ctx, db, s := fixture.ctx, fixture.db, fixture.store-- const advisoryLockID = 9901- lockConn, err := db.Conn(ctx)- require.NoError(t, err)- defer lockConn.Close()- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_lock($1)", advisoryLockID)- require.NoError(t, err)- lockReleased := false- defer func() {- if !lockReleased {- _, _ = lockConn.ExecContext(context.Background(), "SELECT pg_advisory_unlock($1)", advisoryLockID)- }- }()-- _, err = db.ExecContext(ctx, `- CREATE FUNCTION block_task_delete_after_higher_lock() RETURNS trigger AS $$- BEGIN- PERFORM id FROM task- WHERE project = 'project-a' AND id = 102- FOR UPDATE;- PERFORM pg_advisory_xact_lock(9901);- RETURN NULL;- END;- $$ LANGUAGE plpgsql;- CREATE TRIGGER block_task_delete_after_higher_lock- BEFORE DELETE ON task- FOR EACH STATEMENT EXECUTE FUNCTION block_task_delete_after_higher_lock();- `)- require.NoError(t, err)-- deleteResult := make(chan error, 1)- go func() {- deleteResult <- s.DeleteProject(ctx, "default", "project-a")- }()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- SELECT EXISTS (- SELECT 1- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- )- `, advisoryLockID).Scan(&waiting)- return err == nil && waiting- }, 10*time.Second, 10*time.Millisecond, "project deletion should lock task 102 before deleting the task batch")-- operationResult := make(chan error, 1)- go func() {- operationResult <- s.CreatePendingTaskRuns(ctx, "",- &store.TaskRunMessage{ProjectID: "project-a", TaskUID: 101},- &store.TaskRunMessage{ProjectID: "project-a", TaskUID: 102},- )- }()- require.Eventually(t, func() bool {- var waiting bool- err := db.QueryRowContext(ctx, `- WITH delete_backend AS (- SELECT pid- FROM pg_locks- WHERE locktype = 'advisory'- AND objid = $1- AND NOT granted- LIMIT 1- )- SELECT EXISTS (- SELECT 1- FROM pg_stat_activity AS activity, delete_backend- WHERE delete_backend.pid = ANY(pg_blocking_pids(activity.pid))- )- `, advisoryLockID).Scan(&waiting)- return err == nil && waiting- }, 10*time.Second, 10*time.Millisecond, "task run creation should wait for the project purge")-- _, err = lockConn.ExecContext(ctx, "SELECT pg_advisory_unlock($1)", advisoryLockID)- require.NoError(t, err)- lockReleased = true- require.NoError(t, <-deleteResult)- require.NoError(t, <-operationResult)-}--func TestCreatePendingTaskRunsRejectsDeletedProjectDuringProjectDeletion(t *testing.T) {- operationErr, deleteErr := runWithCreationBeforeProjectDeletion(t, `- INSERT INTO instance (resource_id, workspace) VALUES ('instance-a', 'default');- INSERT INTO plan (id, creator, project, name, description) VALUES (101, '[email protected]', 'project-a', 'Plan A', '');- INSERT INTO task (id, project, plan_id, instance, type) VALUES (101, 'project-a', 101, 'instance-a', 'DATABASE_SCHEMA_UPDATE');- `, "task", func(ctx context.Context, s *store.Store) error {- return s.CreatePendingTaskRuns(ctx, "", &store.TaskRunMessage{- ProjectID: "project-a",- TaskUID: 101,- })- })- require.NoError(t, deleteErr)- require.Equal(t, common.NotFound, common.ErrorCode(operationErr))+ return &storePostgresFixture{ctx: ctx, db: db, store: s}}diff --git a/backend/tests/saved_query_test.go b/backend/tests/saved_query_test.goindex 31a9787d37..046225cedf 100644--- a/backend/tests/saved_query_test.go+++ b/backend/tests/saved_query_test.go@@ -786,8 +786,8 @@ func TestCreateSavedQueryRejectsArchivedProject(t *testing.T) {}))a.NoError(err)- // Resource resolution returns archived projects, so without an explicit- // check the create would reach the store's purge fence and read as a 500.+ // Resource resolution returns archived projects, so reject them before the+ // store write._, err = ctl.savedQueryServiceClient.CreateSavedQuery(ctx, connect.NewRequest(&v1pb.CreateSavedQueryRequest{Parent: projectResp.Msg.Name,SavedQuery: &v1pb.SavedQuery{diff --git a/docs/adr/0002-support-workspace-and-project-instances.md b/docs/adr/0002-support-workspace-and-project-instances.mdindex 4d0bc43e66..270b02c212 100644--- a/docs/adr/0002-support-workspace-and-project-instances.md+++ b/docs/adr/0002-support-workspace-and-project-instances.md@@ -176,16 +176,17 @@ blocks data and workflow operations for every caller, including workspace Adminsand DBAs; workspace-level authority does not override an archived parent'slifecycle state. Explicit project restore and purge operations remain available.-Creating a project instance requires its owning project to remain active through-commit. Once the instance exists, in-flight writers beneath it may finish after-the project is archived; their results remain unavailable until the project is-restored. This deliberately avoids making project archival a transaction fence-for descendant metadata.+Creating a project instance checks that its owning project is active. Once the+instance exists, in-flight writers beneath it may finish after the project is+archived; their results remain unavailable until the project is restored. This+deliberately avoids making project archival a transaction fence for descendant+metadata.Permanently purging a project permanently deletes its project instances,databases, and related Bytebase metadata. They are not converted to workspace-instances or moved to the default project. Purge serializes against descendant-writers so that no writer can commit after the owning project is removed.+instances or moved to the default project. Purge does not serialize against+descendant writers; a rare concurrent write may fail with a PostgreSQL conflict+or commit based on state observed before the purge.Directly archiving a project instance leaves its databases assigned to theowning project, but they are unavailable to data and workflow operations until@@ -194,8 +195,9 @@ any targeting task run is `PENDING`, `AVAILABLE`, or `RUNNING`; operators mustcancel that work or wait for it to finish before retrying the lifecycle request.Applying the same guard to restoration prevents legacy queued work fromsilently resuming. Archival does not stop or delete the physical databases.-Purging serializes against descendant writers, then deletes the databases,-their metadata and history, and instance-targeting tasks and task runs.+Purging does not serialize against descendant writers. It deletes the databases,+their metadata and history, and instance-targeting tasks and task runs; a rare+concurrent write may fail or commit from an earlier observed state.Project-level issues, plans, and releases remain, retaining canonical targetnames that may later identify a replacement resource if the purged instance IDis reused; audit logs also remain under the workspace retention policy. Thisdiff --git a/docs/design/saved-queries-model.md b/docs/design/saved-queries-model.mdindex 1ea7545df9..6e942a7a73 100644--- a/docs/design/saved-queries-model.md+++ b/docs/design/saved-queries-model.md@@ -749,73 +749,39 @@ saved_query)` — still group references — indexedper-principal *ordered* seeks (`O(K · page)`) and isolating the ACL indexfrom content churn. A pure storage change behind the same API.-**Lifecycle vs. project deletion.** `saved_query.project` is an FK with no-cascade, so the two directions are fenced explicitly (AGENTS.md transaction-lock-ordering):+**Writer behavior.** `saved_query.project` is an FK with no cascade.-- **Create requires an *active* project.** `CreateSavedQuery` does not go+- **Create requires an active project.** `CreateSavedQuery` does not gothrough `nextProjectID` (IDs are `gen_random_uuid()`, not a project-scoped- sequence), so it must itself, in the insert transaction, lock the project- row and reject the write unless the project is active — a create racing a- purge fails cleanly (`FAILED_PRECONDITION`), never as an FK violation and- never leaving a row in a project mid-deletion.-- **Star writes are child-before-parent; the parent fence covers only the- first star.** The lock rule treats an upsert as an existing-row lock, so- toggling or removing an **existing** `saved_query_star` row locks that- child row directly — no parent lock. Only the **first** star for a- (query, caller) inserts a child that cannot be locked in advance; that- case alone takes the parent fence (lock the `saved_query` row, reject as- `NotFound` if gone), the same missing-child carve-out `CreateSavedQuery`- uses for `project`. Its inserted key is novel, so it never contends with- purge's existing-child locks — no cross-order deadlock.+ sequence), so it verifies the project before inserting.+- **Star writes are child-before-parent.** The lock rule treats an upsert as+ an existing-row lock, so toggling or removing an **existing**+ `saved_query_star` row locks that child row directly. A first star verifies+ its parent saved query before inserting the new child.- **Batch folder moves lock rows in primary-key order.**`MoveMySavedQueries` only *updates* existing `saved_query` rows (nonew child), but "existing" is not enough on its own: two overlappingbatches could grab their target rows in different scan orders anddeadlock. So it locks its selected rows in full primary-key order- (`resource_id`) — the AGENTS.md batch rule — and a folder move touching a- row mid-purge simply updates zero rows.+ (`resource_id`) — the AGENTS.md batch rule.- **Delete is child-before-parent, explicitly.** `DeleteSavedQuery` removesthe query's `saved_query_star` rows before the `saved_query` row itself —- **not** via the FK cascade, which would lock the parent first — matching- the star and purge order so a delete racing a star toggle or a purge- cannot deadlock.-- **Purge is child-before-parent, and re-parents human rows.** The- hard-delete path deletes the stars and saved queries of project service- accounts and workload identities (stars first), then reassigns the- remaining, human-created saved queries to the default project — locking- existing child rows ahead of their parents, per the rule. Because a- surviving row changes project mid-purge, **every non-purge writer scopes- its predicate by `(resource_id, project)`** — patch, delete, and the- first-star parent fence — so a write authorized in the purged project- cannot land on the reassigned row; it updates zero rows and surfaces as- NotFound (the fenced delete rolls back its star cleanup). The declared- lifecycle policy: writers on existing rows require the row in the- authorized project, not an active project — archived projects are already- unreachable through every read path, so a write racing an archive is an- ordinary serialization of concurrent requests, and restore does not- promise to resurrect a row whose authorized delete won that race. Only- creation requires an active project.+ **not** via the FK cascade, which would lock the parent first — so a delete+ racing a star toggle cannot deadlock.++Project and instance archive and physical purge are best-effort lifecycle+operations. A rare concurrent writer may fail with a PostgreSQL conflict or+commit based on state it observed before the lifecycle transition. These+outcomes are accepted and the user can retry the lifecycle operation or write.Two invariants cover every writer. **(1) Child before parent** for-existing-row writes/deletes; the *only* parent-first step is a new-child-*insert* (create → lock `project`; first star → lock `saved_query`), safe-because its key cannot be locked in advance — the AGENTS.md missing-child-carve-out. **(2) Any multi-row lock/update/delete acquires its rows in full-primary-key order** — `saved_query` by `resource_id`, `saved_query_star` by-`(saved_query, principal)` — so two operations over an overlapping set-(delete vs. purge deleting the same query's stars; two batch folder moves;-purge's own `saved_query` batch) can never lock in opposing orders. A plain-`DELETE … WHERE` does not guarantee that order, so these paths take their-row locks through an ordered `… ORDER BY <pk> FOR UPDATE` (or an ordered-delete) before mutating. Required before-implementation: deterministic real-PostgreSQL regression tests for **both**-acquisition orders of each contending pair — create↔purge,-first-star↔purge, delete↔star-toggle, delete↔purge over a query with-multiple stars, and two overlapping `MoveMySavedQueries` over-the same folder — asserting the terminal outcomes — project (or query) deleted, no orphaned-saved query or star, and **no** FK failure or deadlock (`40P01`) in either-direction (absence of `40P01` alone is insufficient).+existing-row writes and deletes. **(2) Any multi-row lock, update, or delete+acquires its rows in full primary-key order** — `saved_query` by+`resource_id`, `saved_query_star` by `(saved_query, principal)` — so+overlapping ordinary operations, such as a delete and star toggle or two+folder moves, cannot lock in opposing orders. A plain `DELETE … WHERE` does+not guarantee that order, so these paths take their row locks through an+ordered `… ORDER BY <pk> FOR UPDATE` (or an ordered delete) before mutating.### Sharing and organization UXdiff --git a/docs/design/sheet-blob-scoping.md b/docs/design/sheet-blob-scoping.mdindex 5e1f94a60b..3d6fcaa1ae 100644--- a/docs/design/sheet-blob-scoping.md+++ b/docs/design/sheet-blob-scoping.md@@ -344,7 +344,7 @@ func (s *Store) GetSheetForProject(ctx, projectID, sha256Hex string, raw bool) (func (s *Store) MissingSheetsForProject(ctx, projectID string, sha256Hexes ...string) ([]string, error)// Scoped creation: writes blobs and ref rows, both as set inserts, in one-// transaction behind the project purge fence.+// transaction after checking the project is active.func (s *Store) CreateSheets(ctx, projectID string, creates ...*SheetMessage) ([]*SheetMessage, error)// Unexported internals: filterSheetsForProject (the ref query),@@ -435,9 +435,8 @@ path.The `project` foreign key makes `sheet_blob_ref` a dependent of `project`, and purge ends with`DELETE FROM project WHERE resource_id = ?` (`backend/store/project.go:826`) after clearing everyother dependent table. Add `DELETE FROM sheet_blob_ref WHERE project = ?` after the `db` delete-(`:704`) and before `project_webhook` (`:739`), matching the position established under-[lock ordering](#transaction-lock-ordering). Without it, purging any project fails on the foreign-key.+(`:704`) and before `project_webhook` (`:739`). Without it, purging any project fails on the+foreign key.Blobs whose only ref belonged to the purged project are left with zero refs, consistent with today'sbehavior — nothing has ever deleted from `sheet_blob` — and the state a future GC would collect.@@ -558,35 +557,15 @@ A large count means either widespread project purges, reused project IDs, or a previsions created without provenance — each worth understanding, since all of them end inunreadable statements.-## Transaction lock ordering+## Transactions-`sheet_blob_ref` is a project-owned branch and takes a position in the canonical sibling order in-`backend/store/AGENTS.md`, updating that list, `DeleteProject`, and `DeleteInstance` together. It is-a direct child of `project` with no descendants, and every table that can reference a hash sits-earlier, so it belongs immediately after `db` and before `project_webhook` — the same position as-the purge delete.--```text-... -> changelog -> sync_history -> revision -> db_schema -> db--> sheet_blob_ref -> project_webhook -> service_account -> ...-```--`CreateSheets` becomes a transaction spanning `sheet_blob` and `sheet_blob_ref`. Insert the blob-first: the ref's foreign key requires it, so the order is forced. Both are `ON CONFLICT DO NOTHING`-— new-row-only inserts, not upserts that can update an existing row, so the README's rule 4 upsert-clause does not apply.--As a writer of purge-managed data, `CreateSheets` must declare a lifecycle policy. **It requires an-active project**: a sheet ref is a new resource with no deleted-project continuation case. The API-create paths check the project before calling the store, but that check is not serialized with-purge, so a concurrent purge would surface as a raw foreign-key violation rather than a controlled-NotFound. `CreateSheets` therefore takes the same transaction-scoped purge fence that database-creation and task-run creation already take; `withDatabasePurgeFence`-(`backend/store/database.go:798-820`) is the pattern to mirror.--Both need the deterministic real-PostgreSQL regression tests that section mandates, asserting-terminal outcomes in both lock-acquisition directions rather than merely the absence of SQLSTATE-`40P01`.+`CreateSheets` inserts blobs before their `sheet_blob_ref` rows because the ref's foreign key+requires it. Both inserts are `ON CONFLICT DO NOTHING` and the transaction verifies that the project+is active before writing. Project and instance archive and physical purge are+best-effort lifecycle operations: a rare concurrent writer may fail with a+PostgreSQL conflict or commit based on state it observed before the lifecycle+transition. These outcomes are accepted and the user can retry the lifecycle+operation or write.## Independent fixes@@ -644,7 +623,7 @@ would flood this query with expected rows. They have their own list above.## TestsIntegration coverage is four tests in `backend/tests/sheet_scope_test.go`, one per independent-decision, plus the migration matrix and the lock-order pair:+decision, plus the migration matrix:- **Cross-project read and cache ordering** — `TestSheetProjectScope`: a sheet created in project Ais readable there (raw and truncated), NotFound under project B — with the raw foreign read@@ -670,9 +649,6 @@ decision, plus the migration matrix and the lock-order pair:exact ref-set match), the stamp outcomes per scenario, the three audits, and a probe perabort-hazard guard (dangling blob, ghost project, bigint overflow, stale stamp, JSON null andmalformed array shapes).-- **Create versus purge** — the deterministic lock-order pair mandated by- `backend/store/AGENTS.md`: `CreateSheets` concurrent with a purge of its project ends in a- controlled NotFound, not a foreign-key error, in both lock-acquisition directions.- **Gate confinement** — a static AST test asserts nothing under `backend/api/v1/` calls`GetSheetFull`; the rest of the split is compiler-enforced by unexported store internals.diff --git a/docs/pre-pr-checklist.md b/docs/pre-pr-checklist.mdindex b9892907d9..2f8ddd8c72 100644--- a/docs/pre-pr-checklist.md+++ b/docs/pre-pr-checklist.md@@ -201,29 +201,16 @@ Read the canonical [store row-lock ordering](../backend/store/AGENTS.md#transact- Transaction-scoped advisory locks are acquired before row locks- Existing related rows are locked child-to-parent- Batches are locked in full primary-key order-- Project-owned sibling branches follow the documented `DeleteProject` order- `nextProjectID` is called only after required existing-child locks- `nextProjectID` locks the project and requires it to be active before allocation;missing or deleted projects reject creation- `UPDATE`, `DELETE`, foreign-key checks, and conflicting upserts are included in the ordering analysis-Row ordering prevents wait-for cycles on existing rows, but it cannot protect an-absent child row. The `nextProjectID` active-project check covers only writers that-call it, not every repository writer. For every new or modified writer of-purge-managed data, define whether it requires an active project or merely an-existing project, then serialize and validate that lifecycle policy against-project deletion.--If the transaction coordinates multiple rows or tables or races with project-deletion, add deterministic real-PostgreSQL regression tests for both-lock-acquisition directions. The tests must observe public store behavior, fail-against the old behavior, and assert terminal outcomes. Verify that neither-direction ends in a foreign-key failure; merely checking for the absence of-SQLSTATE `40P01` is insufficient.+If the transaction coordinates multiple rows or tables, add focused regression+tests for credible ordinary contention paths and assert terminal outcomes.**STOP — do not proceed to PR creation if the transaction conflicts with the-canonical order or lacks the required deterministic lock and lifecycle regression-tests.**+documented order or lacks required regression coverage.**## 6. Image Compatibility Window