Make a call to eth_getblockbynumber with finalized as the block tag. Then, roll the full node back to that block value + 1 using debug_setHead.
Resync the full node from a backup prior to block 12572500, or from genesis.
This issue is resolved and we will be sharing a full post-mortem when ready.
We have identified the root cause of this issue.
The L1 Goerli testnet experienced a large reorg of approximately 18 blocks. In addition to the large reorg, a significant number of blocks were missed, resulting in one large reorg and many missing blocks. This caused a large (>256 block) reorg on OP Goerli. Currently, users operating full nodes on OP Goerli will experience a node halt. These nodes will not sync new blocks from the OP Goerli network, and may return inconsistent data over RPC. Users operating archival nodes on OP Goerli are not affected and we do not expect this issue to occur in any capacity on OP Mainnet.
To resolve this issue, full node operators need to perform one of these actions: