Please solve your RocksDB repairer context switch overhead or make it parallel.

It is currently used in Openethereum.

ssd speed considerably increased since the time it was designed. In my case, I’m seeing currently use 100% cpu on 1 thread and spend 20% in kernel time under Linux. This results in a constant read speed of sst files at 700Mb/s on my Xeon 6250L well below the capacity of my ssd and without switching back to rust code (it’s the same function in the Rocksdb library that executes during hours).

If not improving cpu latency, please at least make the recovery parallel.