> Outcome (added when this file was published, 2026-11-13): fixed: the erasure transaction

From mnema-talk v3.3.13, rendered on this site.

H-L10-06, lane 10 (found in the integration review), S4, reproduced by the reviewer through the real routes on the collection branch and on f58bc3fc6 (3.0.9); no fix in this run

Claim: an account erasure can be aborted by a write of the same member's second device: the erasure locks the users row and later the member's messages; an edit from another device locks the message and then waits on the users row. If the edit has waited more than a second when the erasure reaches its messages step, PostgreSQL detects the deadlock and aborts the erasure: the member's "delete my account" answers 500.

Property: SECURITY.md: account deletion removes the account in one transaction; the H-L10-05 review assumed the erasure always wins.

Repro: an integration test through the real routes with the erasure held by a pgx tracer right before its messages step, written by the integration review and described there (doc/security/reviews/2026-11-hunt/bundle-integration.md, part C, "One cycle that is already on 3.0.9"); the probe itself is not in the repository.

Effect: nothing leaks and nothing is half erased (the transaction rolls back); a second attempt works. Availability of one's own erasure under one's own concurrent write.

Suggested: retry the erasure once or twice on SQLSTATE 40P01. Not done in this run: recorded for the owner.

Something wrong on this page? Tell us.