28316

When a piece of state is flushed from memory to disk, a read request may happen to see it in neither place. Reth PR #28316 describes this handoff race: a historical query first obtains a consistent view, but drops its in-memory snapshot and then reads from the live overlay; if persistence and eviction happen in between, a latest-state query may fail. The fix makes the overlay use the same snapshot. In my view, “it exists on disk and in memory” isn’t enough to ensure reliable queries; a single read must also see the same point in time. This is a fix pending review for the $ETH client. The original text concerns eth_call and eth_estimateGas; these read failures shouldn’t be interpreted directly as assets disappearing on-chain.