It can take a long time to run log recovery operation because it is
single threaded and is bound by read latency. We can find that it took
most of the time to wait for the read IO to occur, so if one object
readahead is introduced to log recovery, it will obviously reduce the
log recovery time.