fix(page): add cycle/depth guard to recursive tree-traversal CTEs (#207)

getPageBreadCrumbs (ancestor CTE) and forceDelete (descendant CTE) used
withRecursive + unionAll with no CYCLE clause or depth cap. If a parent/child
cycle already exists in the data (e.g. one slipped in via the #7 TOCTOU race),
both queries loop forever — hang / statement timeout. Worse, the move guard
itself runs the ancestor CTE, so a cycle would disable the very guard meant to
prevent it (#207 #8).

Add a depth counter bounded by MAX_PAGE_TREE_DEPTH to both recursive CTEs; the
walk stops at the cap, so a cycle yields a bounded result instead of hanging.
Real page trees are only a few levels deep, so the cap never truncates a
legitimate result. getPageBreadCrumbs selects an explicit column list (not
selectAll) so the internal depth counter never leaks into the breadcrumb shape.

Adds an integration test that seeds an A<->B cycle directly and asserts both
getPageBreadCrumbs and forceDelete return bounded / complete under a short
connection-level statement_timeout instead of hanging.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
claude code agent 227
2026-06-26 05:48:01 +03:00
parent 00ca4ff3d6
commit c4807022f2
2 changed files with 169 additions and 4 deletions

View File

@@ -62,6 +62,16 @@ import {
agentSourceFields,
} from '../../../common/decorators/auth-provenance.decorator';
// Hard upper bound on how deep the recursive page-tree CTEs (ancestor /
// descendant traversals) may walk. Real page trees are only a handful of levels
// deep, so this cap never truncates a legitimate result; it purely defends the
// recursive CTEs against runaway iteration if a parent/child cycle ever exists
// in the data (e.g. one slipped in before the move guard, #207 #8). Without it a
// cycle makes `withRecursive` loop forever (hang / statement timeout), and the
// move guard itself calls one of these CTEs — so a cycle would disable the very
// guard meant to prevent it. Each CTE carries a depth counter and stops here.
const MAX_PAGE_TREE_DEPTH = 10_000;
// Advisory-lock namespace (the first key of pg_advisory_xact_lock) used to
// serialize concurrent page moves within a single space so the cycle check and
// the move UPDATE stay atomic (see movePage, #207 #7). A dedicated namespace
@@ -1030,6 +1040,9 @@ export class PageService {
'spaceId',
'deletedAt',
])
// Depth counter: bounds the walk so a parent/child cycle in the data
// can't make this recursive CTE loop forever (#207 #8).
.select(sql<number>`0`.as('depth'))
.where('id', '=', childPageId)
.where('deletedAt', 'is', null)
.unionAll((exp) =>
@@ -1045,12 +1058,25 @@ export class PageService {
'p.spaceId',
'p.deletedAt',
])
.select(sql<number>`pa.depth + 1`.as('depth'))
.innerJoin('page_ancestors as pa', 'pa.parentPageId', 'p.id')
.where('p.deletedAt', 'is', null),
.where('p.deletedAt', 'is', null)
.where(sql<number>`pa.depth`, '<', MAX_PAGE_TREE_DEPTH),
),
)
.selectFrom('page_ancestors')
.selectAll('page_ancestors')
// Explicit column list (not selectAll) so the internal `depth` counter
// never leaks into the breadcrumb result shape.
.select([
'id',
'slugId',
'title',
'icon',
'position',
'parentPageId',
'spaceId',
'deletedAt',
])
.select((eb) =>
eb
.exists(
@@ -1171,16 +1197,21 @@ export class PageService {
db
.selectFrom('pages')
.select(['id'])
// Depth counter: bounds the walk so a parent/child cycle in the data
// can't make this recursive CTE loop forever (#207 #8).
.select(sql<number>`0`.as('depth'))
.where('id', '=', pageId)
.unionAll((exp) =>
exp
.selectFrom('pages as p')
.select(['p.id'])
.innerJoin('page_descendants as pd', 'pd.id', 'p.parentPageId'),
.select(sql<number>`pd.depth + 1`.as('depth'))
.innerJoin('page_descendants as pd', 'pd.id', 'p.parentPageId')
.where(sql<number>`pd.depth`, '<', MAX_PAGE_TREE_DEPTH),
),
)
.selectFrom('page_descendants')
.selectAll()
.select(['id'])
.execute();
const pageIds = descendants.map((d) => d.id);