On November 13 around 7am PST the blockchain halted due to an out of memory bug. All the consensus members, having agreed on the next block’s contents were unable to verify them. This was because, for performance reasons, the transactions in the block were being verified in parallel. Unfortunately there was no limit on the amount of concurrency used and this led to CPU and memory exhaustion as all 65 transactions in block 110,710 were being verified simultaneously.
A hotfix to limit the number of transactions verified at once was developed and rolled out around 10:30am PST to the consensus group. This immediately allowed the in-progress block to be completed and gossipped around. However non-concensus nodes without the fix, when trying to validate the new block, then began to OOM as they saw the new block. An OTA build was completed around 11:20am PST and deployed to the beta group. Shortly afterwards, assuming everything looks good, we will do a deployment to the wider network.
Longer term, this relates to the rapid growth of the network. We are still working hard on improving the scalability of the systems as the network grows and expect to deploy significant improvements to make transactions much cheaper to validate soon.
- Limit pmap width when validating signatures and transactions: PRs blockchain-core/#292
The OTA is already in beta and will be deployed more widely soon.