Mirroristas/etcd
2
0
Fork 0
mirror of https://github.com/etcd-io/etcd.git synced 2024-09-27 06:25:44 +00:00
Code Issues Packages Projects Releases Wiki Activity

*: fix spelling issues (codespell).

Browse Source 
Signed-off-by: Dmitry Smirnov <onlyjob@member.fsf.org>
This commit is contained in:
Dmitry Smirnov
2015-09-11 08:06:56 +10:00
parent 1de63deca4
commit b2f4a5f587
15 changed files with 17 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
raft/design.md
View File

@@ -45,7 +45,7 @@ When in `snapshot` state, leader stops sending any `replication message`.
A newly elected leader sets the progresses of all the followers to `probe` state with `match` = 0 and `next` = last index. The leader slowly (at most once per heartbeat) sends `replication message` to the follower and probes its progress.
A progress changes to `replicate` when the follower replies with a non-rejection `msgAppResp`, which implies that it has matched the index sent. At this point, leader starts to stream log entries to the follower fast. The progress will fall back to `probe` when the follower replies a rejection `msgAppResp` or the link layer reports the follower is unreachable. We aggressively reset `next` to `match`+1 since if we receive any `msgAppResp` soon, both `match` and `next` will increase directly to the `index` in `msgAppResp`. (We might end up with sending some duplicate entries when agressively reset `next` too low. see open question)
A progress changes to `replicate` when the follower replies with a non-rejection `msgAppResp`, which implies that it has matched the index sent. At this point, leader starts to stream log entries to the follower fast. The progress will fall back to `probe` when the follower replies a rejection `msgAppResp` or the link layer reports the follower is unreachable. We aggressively reset `next` to `match`+1 since if we receive any `msgAppResp` soon, both `match` and `next` will increase directly to the `index` in `msgAppResp`. (We might end up with sending some duplicate entries when aggressively reset `next` too low. see open question)
A progress changes from `probe` to `snapshot` when the follower falls very far behind and requires a snapshot. After sending `msgSnap`, the leader waits until the success, failure or abortion of the previous snapshot sent. The progress will go back to `probe` after the sending result is applied.