fix: avoid deleting peers on graceful close (#14165)

* fix: avoid deleting peers on graceful close

- Fixes an issue where a coordinator deletes all
  its peers on shutdown. This can cause disconnects
  whenever a coderd is redeployed.

coderd/database/dbauthz/dbauthz.go

@@ -3324,6 +3324,13 @@ func (q *querier) UpdateReplica(ctx context.Context, arg database.UpdateReplicaP
 	return q.db.UpdateReplica(ctx, arg)
 }
 
+func (q *querier) UpdateTailnetPeerStatusByCoordinator(ctx context.Context, arg database.UpdateTailnetPeerStatusByCoordinatorParams) error {
+	if err := q.authorizeContext(ctx, policy.ActionUpdate, rbac.ResourceTailnetCoordinator); err != nil {
+		return err
+	}
+	return q.db.UpdateTailnetPeerStatusByCoordinator(ctx, arg)
+}
+
 func (q *querier) UpdateTemplateACLByID(ctx context.Context, arg database.UpdateTemplateACLByIDParams) error {
 	fetch := func(ctx context.Context, arg database.UpdateTemplateACLByIDParams) (database.Template, error) {
 		return q.db.GetTemplateByID(ctx, arg.ID)

coderd/database/dbauthz/dbauthz_test.go

@@ -2152,6 +2152,11 @@ func (s *MethodTestSuite) TestTailnetFunctions() {
 			Asserts(rbac.ResourceTailnetCoordinator, policy.ActionCreate).
 			Errors(dbmem.ErrUnimplemented)
 	}))
+	s.Run("UpdateTailnetPeerStatusByCoordinator", s.Subtest(func(_ database.Store, check *expects) {
+		check.Args(database.UpdateTailnetPeerStatusByCoordinatorParams{}).
+			Asserts(rbac.ResourceTailnetCoordinator, policy.ActionUpdate).
+			Errors(dbmem.ErrUnimplemented)
+	}))
 }
 
 func (s *MethodTestSuite) TestDBCrypt() {

coderd/database/dbmem/dbmem.go

@@ -7917,6 +7917,10 @@ func (q *FakeQuerier) UpdateReplica(_ context.Context, arg database.UpdateReplic
 	return database.Replica{}, sql.ErrNoRows
 }
 
+func (*FakeQuerier) UpdateTailnetPeerStatusByCoordinator(context.Context, database.UpdateTailnetPeerStatusByCoordinatorParams) error {
+	return ErrUnimplemented
+}
+
 func (q *FakeQuerier) UpdateTemplateACLByID(_ context.Context, arg database.UpdateTemplateACLByIDParams) error {
 	if err := validateDatabaseType(arg); err != nil {
 		return err

coderd/database/dbmetrics/dbmetrics.go

@@ -2069,6 +2069,13 @@ func (m metricsStore) UpdateReplica(ctx context.Context, arg database.UpdateRepl
 	return replica, err
 }
 
+func (m metricsStore) UpdateTailnetPeerStatusByCoordinator(ctx context.Context, arg database.UpdateTailnetPeerStatusByCoordinatorParams) error {
+	start := time.Now()
+	r0 := m.s.UpdateTailnetPeerStatusByCoordinator(ctx, arg)
+	m.queryLatencies.WithLabelValues("UpdateTailnetPeerStatusByCoordinator").Observe(time.Since(start).Seconds())
+	return r0
+}
+
 func (m metricsStore) UpdateTemplateACLByID(ctx context.Context, arg database.UpdateTemplateACLByIDParams) error {
 	start := time.Now()
 	err := m.s.UpdateTemplateACLByID(ctx, arg)

coderd/database/dbmock/dbmock.go

@@ -4366,6 +4366,20 @@ func (mr *MockStoreMockRecorder) UpdateReplica(arg0, arg1 any) *gomock.Call {
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "UpdateReplica", reflect.TypeOf((*MockStore)(nil).UpdateReplica), arg0, arg1)
 }
 
+// UpdateTailnetPeerStatusByCoordinator mocks base method.
+func (m *MockStore) UpdateTailnetPeerStatusByCoordinator(arg0 context.Context, arg1 database.UpdateTailnetPeerStatusByCoordinatorParams) error {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "UpdateTailnetPeerStatusByCoordinator", arg0, arg1)
+	ret0, _ := ret[0].(error)
+	return ret0
+}
+
+// UpdateTailnetPeerStatusByCoordinator indicates an expected call of UpdateTailnetPeerStatusByCoordinator.
+func (mr *MockStoreMockRecorder) UpdateTailnetPeerStatusByCoordinator(arg0, arg1 any) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "UpdateTailnetPeerStatusByCoordinator", reflect.TypeOf((*MockStore)(nil).UpdateTailnetPeerStatusByCoordinator), arg0, arg1)
+}
+
 // UpdateTemplateACLByID mocks base method.
 func (m *MockStore) UpdateTemplateACLByID(arg0 context.Context, arg1 database.UpdateTemplateACLByIDParams) error {
 	m.ctrl.T.Helper()

coderd/database/dbtestutil/db.go

@@ -35,6 +35,7 @@ type options struct {
 	dumpOnFailure bool
 	returnSQLDB   func(*sql.DB)
 	logger        slog.Logger
+	url           string
 }
 
 type Option func(*options)
@@ -59,6 +60,12 @@ func WithLogger(logger slog.Logger) Option {
 	}
 }
 
+func WithURL(u string) Option {
+	return func(o *options) {
+		o.url = u
+	}
+}
+
 func withReturnSQLDB(f func(*sql.DB)) Option {
 	return func(o *options) {
 		o.returnSQLDB = f
@@ -92,6 +99,9 @@ func NewDB(t testing.TB, opts ...Option) (database.Store, pubsub.Pubsub) {
 	ps := pubsub.NewInMemory()
 	if WillUsePostgres() {
 		connectionURL := os.Getenv("CODER_PG_CONNECTION_URL")
+		if connectionURL == "" && o.url != "" {
+			connectionURL = o.url
+		}
 		if connectionURL == "" {
 			var (
 				err     error

coderd/database/querier.go

@@ -418,6 +418,7 @@ type sqlcQuerier interface {
 	UpdateProvisionerJobWithCancelByID(ctx context.Context, arg UpdateProvisionerJobWithCancelByIDParams) error
 	UpdateProvisionerJobWithCompleteByID(ctx context.Context, arg UpdateProvisionerJobWithCompleteByIDParams) error
 	UpdateReplica(ctx context.Context, arg UpdateReplicaParams) (Replica, error)
+	UpdateTailnetPeerStatusByCoordinator(ctx context.Context, arg UpdateTailnetPeerStatusByCoordinatorParams) error
 	UpdateTemplateACLByID(ctx context.Context, arg UpdateTemplateACLByIDParams) error
 	UpdateTemplateAccessControlByID(ctx context.Context, arg UpdateTemplateAccessControlByIDParams) error
 	UpdateTemplateActiveVersionByID(ctx context.Context, arg UpdateTemplateActiveVersionByIDParams) error

coderd/database/queries.sql.go

@@ -7419,6 +7419,25 @@ func (q *sqlQuerier) GetTailnetTunnelPeerIDs(ctx context.Context, srcID uuid.UUI
 	return items, nil
 }
 
+const updateTailnetPeerStatusByCoordinator = `-- name: UpdateTailnetPeerStatusByCoordinator :exec
+UPDATE 
+	tailnet_peers
+SET
+	status = $2
+WHERE
+	coordinator_id = $1
+`
+
+type UpdateTailnetPeerStatusByCoordinatorParams struct {
+	CoordinatorID uuid.UUID     `db:"coordinator_id" json:"coordinator_id"`
+	Status        TailnetStatus `db:"status" json:"status"`
+}
+
+func (q *sqlQuerier) UpdateTailnetPeerStatusByCoordinator(ctx context.Context, arg UpdateTailnetPeerStatusByCoordinatorParams) error {
+	_, err := q.db.ExecContext(ctx, updateTailnetPeerStatusByCoordinator, arg.CoordinatorID, arg.Status)
+	return err
+}
+
 const upsertTailnetAgent = `-- name: UpsertTailnetAgent :one
 INSERT INTO
 	tailnet_agents (

coderd/database/queries/tailnet.sql

@@ -149,6 +149,14 @@ DO UPDATE SET
 	updated_at = now() at time zone 'utc'
 RETURNING *;
 
+-- name: UpdateTailnetPeerStatusByCoordinator :exec
+UPDATE 
+	tailnet_peers
+SET
+	status = $2
+WHERE
+	coordinator_id = $1;
+
 -- name: DeleteTailnetPeer :one
 DELETE
 FROM tailnet_peers