Skip to content

Retry an after-response write that loses its CAS #1553

Open

Description

@dhruv8sh

Follow-up from PR #1487.

Description

The after-response write re-reads the row and writes once. A conflict, often with this navigation's own pull-sync or another write, drops the results, and the vendor is asked again on the next navigation.

Proposed direction: Measure the conflict rate, then allow one bounded re-read and re-plan on Conflict, keeping the single-permit rule.

Done when

  • The after-response CAS conflict rate is measured (log or metric).
  • On Conflict, one bounded re-read and re-plan is attempted, keeping the single-permit rule.
  • Tests cover a conflict followed by a successful retry and a second conflict that gives up.

Affected area

Core (Edge Cookies, GDPR)

Code: crates/trusted-server-core/src/ec/after_response.rs (commit_after_response_lookups); ec/kv.rs (commit_after_response).

Review threads

Activity

  1. added this to the 202611 milestone on Nov 11, 2026
  2. self-assigned this
    on Nov 11, 2026
  3. added theissue type on Nov 11, 2026
  4. aram356 commented on Nov 13, 2026

    @aram356
    Collaborator

    CREATOR: This issue is now part of #1558, server-side identity on every adapter. It is planned in wave 3 (MVP: Fastly and Axum); priority P2, size S.

    Blocked by:

    Related pull requests:

    Design: spec; order and dependencies: plan.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions