The Tendermint-core uses a write-ahead log (WAL) mechanism to guarantee consensus safety after crashed validators recover. But this mechanism requires validators to write any received or generated messages to be written to the WAL, whose content will be synced to the disk. This process seems to incur heavy disk IO in my computer. I wonder whether the developers of Tendermint have provided any method or mode to remove the heavily used write-ahead log mechanism.
1 Like
I don’t think that any such thing exists.
My understanding of the wal is that it ensures that signatures are always available, should a node crash.
The idea of having a toggle for it is kind of interesting.
Would need to study safety impacts.
How much disk io is wal using in your case?
1 Like
The WAL is necessary for crash-recovery. But it can be optimized. This could be a good place for continuing the discussion—Option to reduce cs.wal max size · Issue #1233 · cometbft/cometbft · GitHub
By the way, Tendermint Core is no longer maintained. Switch to CometBFT.