[consensus] gate slim block emission on subscriber lag

[arun-koshy]

Description

Serves full blocks to a subscriber that has fallen too far behind to rebuild minimal ones, and re-decides as its lag changes. Required before consensus_slim_block_propagation can be turned on.

Reconstruction is all-or-nothing across a block's ~100 ancestors: one unresolvable slot fails the whole block, so a small per-ancestor miss rate compounds into near-total failure once a receiver trails the omission horizon. A lagging cohort measured 76% of blocks failing to rebuild against 16% at the frontier, and sending those blocks wastes both sides' work. Full blocks put such a subscriber back on the ordinary path — verify, suspend, fetch missing ancestors — which is how it catches up.

Thresholds are expressed against gc_depth because that is the horizon reconstruction can resolve within. They are asymmetric: a subscription loses slim blocks at the horizon but earns them back only well inside it. Without that gap a subscriber sitting on a single threshold flapped at ~100 transitions/s in each direction, sustained, while the fleet stalled — each flip changes the wire form mid-stream, so neither mode took effect. A dwell timer bounds how often a subscription may switch at all.

Entry uses the strict threshold, since bootstrap starts every subscriber far from the frontier. The gate then keeps tracking: a node that falls behind mid-stream would otherwise be served slim blocks indefinitely on a decision made while it was healthy — a lagging cohort held a ~450-round deficit while still being sent 347 slim blocks/s, 84% of which it could not rebuild.

Catch-up replay is unchanged and still sends full blocks; it serves rounds where the subscriber's state is unpredictable by construction.

Stacked on #27702 — review that first; this targets its branch.

Test plan

Four tests. The mode decision is extracted to gate_next_mode so the hysteresis is testable without a stream harness: the thresholds leave a real gap (mutation-verified — collapsing it fails the test), the mode holds inside the gap from either direction, and it switches outside it. subscriber_lag is tested against a real DagState to confirm it takes the later of the peer's reported position and its observed one.

With the flag off — every chain today — the gate evaluates once at subscribe time to false and the re-evaluation branch never runs.

---

Release notes

Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required.

For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates.

• Protocol:
• Nodes (Validators and Full nodes):
• gRPC:
• JSON-RPC:
• GraphQL:
• CLI:
• Rust SDK:
• Indexing Framework:

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
sui-docs	Ready	Preview	Aug 21, 2026 9:27pm

2 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
multisig-toolkit	Ignored	Preview	Aug 21, 2026 9:27pm
sui-kiosk	Ignored	Preview	Aug 21, 2026 9:27pm