Revert "consensus: restructure peer catchup sleep" #783
Conversation
This reverts commit a889f17.
|
Why are we reverting tendermint/tendermint#8651? Isn't that a fix? The rationale for this PR isn't documented. Is 8651 being reverting upstream as well? |
As mentioned briefly in the description and in celestiaorg/celestia-app#475, using v0.35.6 with large blocks results in consensus taking many rounds, as block producers will prematurely start a new round as block gossiping is too slow. This PR fixes this by reverting the change, however this PR was created before talking with the tendermint team to find a better solution. it became clear that the default settings also need to be changed in order for the fix to work, specifically |
* Unsafe int cast in `kill` command (#783) * Unsafe int cast in `kill` command * Revert "Unsafe int cast in `kill` command" This reverts commit bbd649bd372ca90f83dea7b424d67dafbd9eb541. * Changed strategy (cherry picked from commit 03c5e7727a03983b54623e731d5d3d8dd4ac75ec) # Conflicts: # cmd/cometbft/commands/debug/kill.go * Revert "Unsafe int cast in `kill` command (#783)" This reverts commit b7ab279a6df1f062bec60bcf95947d2a87f4ccec. * Unsafe int cast in `kill` command (#783) * Unsafe int cast in `kill` command * Revert "Unsafe int cast in `kill` command" This reverts commit bbd649bd372ca90f83dea7b424d67dafbd9eb541. * Changed strategy --------- Co-authored-by: Sergio Mena <sergio@informal.systems>
This reverts commit a889f17.
When proposing large blocks, this refactor to timeouts was contributing to nodes prematurely starting a new round before the block could be gossiped.