That sounds like you are using ZK in a way that is more appropriate for a database. Particularly if you are putting watches on a large fraction of those 40 million znodes.
ZK is not intended to scale up or out. It is intended to serve as a coordination layer for other systems that do. As such what you need to solve your problem is a bit of a redesign so that you *leverage* the strong characteristics of ZK to allow other systems to scale more simply. On Fri, Jul 24, 2020 at 12:05 AM happen <370119...@qq.com> wrote: > Hi All, > > > In test lab, I tested 3 participants quorum that have 40million > znodes and 5GiB snapshot file. > > > it took 8 minutes to make quorum be in serving status after leader shut > down. > > > Do you have any suggestions to make it faster ?
