Skip to content

fix(release): refresh crates observation before authorize after Retry-After waits #325

Description

@DecisionNerd

Objective

Stop crates lane authorize from hard-failing with blocked_registry_state (indeterminate) when a still-absent crate's initial observe-all snapshot ages past MAX_OBSERVATION_AGE_SECONDS (10m) during durable crates.io Retry-After waits.

Evidence

  • Run: https://github.com/CurateLabs/graphforge/actions/runs/30715697365
  • After graphforge-rel accepted (following a 523s Retry-After sleep), authorize for crates:graphforge-search exited 1 with blocked_registry_state (indeterminate).
  • No search attempt/receipt was written.
  • Live crates.io: graphforge-search does not exist (404). Reconciliation for that run classifies search as absent / publish.
  • Root cause: plan_recovery rewrites observations older than 10 minutes to observation_stale → indeterminate. The crates loop refreshes only the crate just published, so remaining absent crates keep the initial observe-all timestamp.

Done when

  • The crates publish loop refreshes the current node's live observation immediately before authorize.
  • A contract test locks that ordering/behavior.
  • Authorize failure text includes the observation reason when disposition is blocked (for operability).
  • Tag v0.5.1 remains at dd1fd8cdeb75aba11948ed6e1871e98a46560be8 (no retag).

Non-goals

  • Moving or retagging v0.5.1.
  • Weakening observation age checks.
  • Thrashing redispatches without this fix on main.

Activity

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions