-
Notifications
You must be signed in to change notification settings - Fork 1.6k
Chain Reversion Load Test #6735
Comments
Was able to get confirmation from traces that reversions are primarily happening via overseer message now. Here are some results from Versi with 211 active validators, 5 of which are producing garbage candidates. Average reversion time after dispute resolution: 97.7% (760/778) of reversions handled via overseer message. Exceptions are when chain selection scrapes a reversion before the dispute coordinator has processed statement imports. This happens when a node is down or isolated from peers while a dispute is being decided. |
Example reversions caused by a single parablock candidate:Parent blocks reverted via chainSelectionMessage::RevertBlock:
ChainSelectionMessage::RevertBlocks:
Block we hadn’t seen is later reverted via On-chain: |
For each relay parent only one of the two reversion methods can be used, since the first reversion method to trigger removes the relevant block_entry in the chain selection tree. |
Awesome, thanks Bradley! |
Run concluding invalid disputes (with reversions) on Versi again to see how well we handle those these days. In particular, it would be good to see this in action, ideally with some data how much quicker we can revert under load with that change.
The text was updated successfully, but these errors were encountered: