FlatDb 1.37.1: restart mid-snap promotes incomplete state, deletes canonical chain, sync permanently stuck

[svlachakis]

Description

After updating to 1.37.1 and syncing into a fresh FlatDb, the node reaches a few hundred blocks behind tip, stops downloading, then on restart deletes thousands of canonical blocks as "invalid" and remains permanently stuck at Downloaded 0 / 25,002,159 (0.00%) with repeated Unable to find beacon header messages on every ForkChoice update.

The deleted blocks are canonical (verified against Etherscan, e.g. block 24996520 = 0xa2a304db…). Recovery requires wiping the DB and resyncing.

This appears related to but distinct from #11418, #11442, and #11447 (all FlatDb / restart-during-snap symptoms in 1.37.1), but the failure mode here is canonical-chain deletion + unrecoverable stall, which makes it more severe.

Steps to Reproduce

1. Sync a Nethermind 1.37.1 node with FlatDb enabled into a fresh DB.
2. Let snap sync progress until close to tip but before completing (snap state-nodes phase still running, e.g. State Sync at < 1%) (user: when it gets close to the tip (few hundred blocks) then doesn't download anything).
3. Stop and restart the container.
4. Observe: on restart the node logs InvalidTransactionException: transaction nonce is too high on the first post-pivot block, then deletes thousands of subsequent canonical blocks as invalid and never recovers.

Desktop

• Operating System: Linux (arm64)
• Version: 1.37.1
• Installation Method: Docker

Additional context

nethermind-laststart.txt
message.txt