Crypto History

The Six Hours Bitcoin Had Two Blockchains

The Six Hours Bitcoin Had Two Blockchains

On 11 March 2013, bitcoin stopped agreeing with itself. For about six hours there were two competing blockchains, each considered valid by a large share of the network, each accepting transactions the other didn't recognize. It is the closest bitcoin has come to simply falling apart, and the cause was not an attacker — it was a routine upgrade.

A quirk nobody knew was a rule

Bitcoin Core version 0.8 replaced the aging Berkeley DB storage layer with LevelDB, a straightforward engineering improvement that made the software faster. What nobody had noticed was that Berkeley DB carried an undocumented limit on the number of locks it could hold while processing a block. That limit had never mattered, because no block had ever been big or complex enough to hit it.

On that day, a mining operation that had upgraded to the new 0.8 software produced a block that was unusually large for the time and touched enough distinct records to exceed the old limit. Nodes running 0.8 processed it without complaint. Nodes running 0.7 and earlier tried, hit the Berkeley DB lock ceiling, and rejected it as invalid.

That is the definition of an accidental hard fork. Half the network built on top of a block the other half refused to acknowledge, and the two chains diverged from there. A transaction confirmed on one chain might not exist at all on the other — which is exactly the double-spend risk bitcoin exists to prevent. At least one double-spend was in fact executed against a payment processor during the window.

The fix nobody could enforce

Developers converged on IRC within minutes and faced a genuinely awkward choice. The 0.8 chain was longer and technically "correct" by the new software's rules. The 0.7 chain was the one the majority of economic nodes, exchanges and older users were on.

They chose to roll back to the old behavior: ask miners running 0.8 to downgrade to 0.7 and abandon the chain they'd been building. That meant voluntarily discarding several hours of valid, mined blocks — real money, forfeited, by request. Large mining pools agreed and switched. The 0.7 chain overtook the 0.8 chain, the network reconverged on a single history, and the incident was over in roughly six hours. Mt. Gox and other exchanges suspended deposits during the episode, and the price dropped sharply before recovering.

The permanent fix came later and more carefully: an agreed deadline after which the old Berkeley DB limit would stop being enforced, so the network could move to 0.8's behavior deliberately rather than by accident.

Why this one still gets cited

Three things make March 2013 the most instructive crisis in bitcoin's technical history.

Consensus rules include the bugs. In a system where every participant must compute the same answer, an accidental quirk of a database library is part of the protocol, whether or not it was ever written down. This reframed how developers think about changes: any modification that could cause two versions to disagree about a block's validity is a consensus change, no matter how internal it looks.

Soft forks over hard forks. The episode is a large part of why bitcoin development strongly prefers backward-compatible upgrades, where old nodes still accept new blocks. Later changes like SegWit and Taproot were engineered as soft forks specifically to avoid a repeat.

Coordination was the actual mechanism. No one could force miners to abandon their blocks. The recovery depended on a small number of operators being reachable, persuadable, and willing to take a loss for the health of the network. That worked in 2013, when the relevant people fit in a chat room. It's a fair question whether it would work as smoothly today — and a preview of the governance fight that would arrive with the block size wars.

← All articles