[Skeyelab]

Problem

Found during the Hermes plugin rollout. A connected node only gets an event when it first arrives (ingress push) or when the node authenticates (pushPendingEventsForNode). If the node doesn't ack (in the rollout, the local Hermes route returned 404), the event stays leased and is never retried while the connection stays open. Two test events sat stuck for 4+ minutes after their 60s lease expired.

Before #81 the relay reconnected every ~45s, which quietly re-sent stuck events on each new connection. The keepalive made connections stable, so the gap is now real.

Change

app/lib/ws/node-socket.ts: on each 30s keepalive tick, call pushPendingEvents(ws, nodeId), the same function used at auth, which selects pending plus lease-expired leased events and re-leases them. So nothing is sent twice inside a lease window. A tick that terminates an unresponsive node doesn't re-send.

Tests (written first)

• re-sends on each keepalive tick
• nothing is re-sent to a node being terminated
• a failed re-send is logged, not thrown

yarn vitest run: 245/245. Lint clean, no type errors outside tests.

[strix-security]

Strix Security Review

No security issues found.

Review summary

The PR adds retry of pending/lease-expired events on each of a node's WebSocket keepalive ticks, reusing the existing pushPendingEvents path rather than introducing new logic. The changed connection handler was reviewed together with the wired dependencies in the server entrypoint. All database interactions remain parameterized and scoped to the authenticated node, no authorization or input-validation controls were removed, the unresponsive-node termination path correctly skips re-sending, and retries are bounded by an attempt count. No security vulnerabilities were identified in the changed code.

_{Updated for 071a018.}

---

Reviewed by Strix
_{Re-run review · Configure security review settings}