TYPO: "..., it could have two different uncommitted log entries in two different machines. ..." should be "..., it could have two different uncommitted log entries having the same index in two different machines. ..."
Tsz-Wo On Sun, Mar 19, 2023 at 1:31 PM Tsz Wo Sze <[email protected]> wrote: > Requests from different clients may arrive at the same time. When the > leader receives client requests, it assigns a unique log index > (monotonically) for each request. The log index decides the serial order > of the client requests. Within a leader, two different requests cannot > have the same index. > > > ... Uncommitted log may have same index, and one of them will be > committed, others will be overwritten. ... > > Suppose there is a leader change and C is the last commit index of the old > leader. The old leader may have assigned log indices C+1, C+2, ... to some > uncommitted requests. The new leader begins assigning log indices to new > client requests at C+1. Then, it could have two different uncommitted > log entries in two different machines. When the old leader joins the new > leader, it will truncate its uncommitted log entries. > > Hope it helps. > > Tsz-Wo > > > On Sun, Mar 19, 2023 at 12:11 AM tison <[email protected]> wrote: > >> Thank you for quick response! >> >> Best, >> tison. >> >> >> Kaijie Chen <[email protected]> 于2023年3月19日周日 00:09写道： >> >> > No. Index of committed log increases monotonically.Uncommitted log may >> > have same index, and one of them will be committed, others will be >> > overwritten.Kaijie ---- On Sun, 19 Mar 2023 00:03:10 +0800 >> > [email protected] wrote ----IIUC different terms can have the same >> > index. I'm not quite familiar with >> > the underneath promise here. >> > >> > The API in issue is org.apache.ratis.client.api.AsyncApi#watch >> > >> > CompletableFuture<RaftClientReply> watch(long index, ReplicationLevel >> > replication); >> > >> > Best, >> > tison. >> > >> >
