Repository navigation
test: add index corruption recovery and rebuild fixtures #215
Description
Source
Extracted from the reliability audit snippet: indexing/storage recovery from index-file corruption and checksum behavior.
Programmatic test scope
Add deterministic corruption fixtures for on-disk indexes and verify recovery/rebuild behavior.
Cases to cover
- Corrupted index checksum with intact fact pages.
- Corrupted B+tree leaf/internal page bytes.
- Root page pointer mismatch or impossible page ID.
- Fact pages intact but one or more indexes unusable.
Acceptance criteria
- Expected rebuild paths are covered by tests.
- Irrecoverable corruption returns a clear error.
- Query results after rebuild match pre-corruption expected facts.
Activity
adityamukho commentedon May 2, 2026
Audit finding: The
index_checksumfield (persistent_facts.rs:842) is misleadingly named — it checksums fact data pages (1 throughnew_total_fact_pages), NOT the B+tree index pages. A bit-flip in any index page (which starts at1 + new_total_fact_pages) won't trigger a rebuild on reopen becauseneeds_rebuild(line 248-249) only checks the fact-page checksum.The corruption fixtures in this issue should include a test case where index pages are corrupted but fact pages are intact — this currently passes silently (no rebuild triggered), producing wrong query results.
chore(wave3): foundation — FaultInjectingBackend, fuzz/ crate, propte…
Merge pull request #255 from project-minigraf/wave3/pr3-storage