Base has defined why its mainnet stopped producing blocks twice in two days.
Abstract
(*26*)
The Coinbase-backed Ethereum layer-2 community mentioned each outages got here from the same bug in its sequencer block-building logic.
The primary outage started on June 25 and lasted about 116 minutes. The second started on June 26 and lasted about 20 minutes. Base mentioned funds stayed protected throughout each incidents.
(*25*)Sequencer bug stopped block manufacturing
In its official postmortem, Base mentioned an invalid transaction failed throughout execution, as anticipated. The difficulty got here after that failure, when stale journal state remained contained in the block builder.
That stale state included accounts and storage slots touched by the failed transaction. When a legitimate transaction got here subsequent, the system used the incorrect journal state and charged gasoline incorrectly.
This created a block with an invalid state transition. Different nodes couldn’t settle for the block, so the chain stopped producing new L2 blocks.
“The integrity of the chain was not compromised and all funds on Base had been protected,” Base mentioned.
The group added that block manufacturing resumed safely after mitigation.
(*25*)Transactions queued throughout the halt
Throughout the outages, customers couldn’t get new transactions included onchain. Base mentioned transactions queued within the mempool whereas the chain waited for block manufacturing to recuperate.
The transaction pool later grew past what it may retailer. In consequence, new eth_sendRawTransaction requests returned errors throughout the outage window.
The halt additionally affected sequencer and validator progress. Base mentioned these nodes couldn’t transfer past the invalid block till sequencing returned.
As previously reported, Base first flagged unhealthy block manufacturing on June 25 earlier than engineers remoted a consensus downside tied to an invalid block.
(*25*)Patch mounted stale state problem
Base mentioned it mounted the principle bug by making use of a sequencer patch. The patch ensures journal state updates correctly throughout execution after a failed transaction.
The group additionally discovered a second problem throughout restoration. Base mentioned mitigation took longer as a result of a race situation within the engine reset function stopped sequencers from catching up after restart.
That second problem helped clarify why the incident returned the following day. Base mentioned the issue affected sequencers, not validator nodes, but it surely nonetheless slowed restoration.
The Base status page confirmed sequencing resumed on June 25. It additionally instructed ecosystem node operators to restart Base nodes in the event that they had been nonetheless caught.
(*25*)Testing and restoration modifications deliberate
Base mentioned it can strengthen protocol fuzz testing and load testing. These strategies assist groups discover unusual transaction patterns which will expose hidden bugs.
The group additionally plans higher monitoring and operational checks. It mentioned these modifications ought to assist engineers detect comparable issues earlier and reply sooner.
Base additionally desires so as to add sleek restoration to base-consensus. That change would make it simpler for validator nodes to proceed syncing after comparable failures.
The outage got here throughout a busy week for the community. Base additionally moved ahead with its Beryl upgrade, which provides the B20 token normal and cuts the usual Base-to-Ethereum withdrawal interval from seven days to 5 days.
The incident offers builders and customers a clearer view of the weak level. Base has now named the bug, launched a patch, and listed the exams it plans to enhance.
