Skip metadata reconciliation after replica demotion - #1304

Included review availability: Your plan includes up to 2 reviews per rolling hour; 1 remains after this review.

Walkthrough

Changes

Metadata demotion guard

Sequence Diagram(s)

sequenceDiagram
  participant AsyncMetadataRetrieval
  participant MetadataManager
  participant EventLoop
  participant ContextFlags
  AsyncMetadataRetrieval->>MetadataManager: deliver fetched metadata
  MetadataManager->>EventLoop: queue reconciliation callback
  ContextFlags->>MetadataManager: mark node as replica
  EventLoop->>MetadataManager: execute queued callback
  MetadataManager->>ContextFlags: check replica role
  MetadataManager-->>MetadataManager: discard reconciliation

Suggested reviewers: boda26

• X
• Mastodon
• Reddit
• LinkedIn

This change prevents a node that has been demoted to a replica from applying metadata fetched asynchronously while it was primary. The queued callback now rechecks the node role before reconciliation, and the added regression coverage models demotion between retrieval and callback dispatch.

Safe to merge: the demotion path retains existing metadata and suppresses reconciliation callbacks.

No blocking failure remains.

What T-Rex did

• Ran the executable metadata demotion regression harness against the base and updated queued metadata callbacks to exercise the demotion flow.
• Compared the base run to the PR-head run; the updated callback places the guard before ReconcileMetadata, resulting in zero reconciliations and zero update callbacks and the existing metadata is retained.
• Used the trex-artifacts/demotion-gate-runtime-harness.py to verify guard placement and exercise the demotion dispatch model against base and PR-head configurations.
• Captured native-test blockers for the coordinator flow, including missing CMake prerequisites and the absence of a compiled coordinator test binary.

_{Ran code and verified through T-Rex}