After a node stops, after deleting the data file, restart the reason for repeatedly printing the following logs. Is there any solution?

sofastack / sofa-jraft

A production-grade java implementation of RAFT consensus algorithm.

https://www.sofastack.tech/projects/sofa-jraft/

Apache License 2.0

3.59k stars 1.15k forks source link

After a node stops, after deleting the data file, restart the reason for repeatedly printing the following logs. Is there any solution? #315

Closed huzhihong closed 5 years ago

huzhihong commented 5 years ago

onSnapshotLoad这个方法在启动的时候会被调用多次，我代码其中有一个逻辑是判断当前状态机有没有数据，已经有数据就不需要从快照恢复，onSnapshotLoad直接返回false，最后导致了结点状态一致不正确。现在我想问onSnapshotLoad被调用多次的情况下，是以快照数据为准，还是以当前状态机的数据为准？

fengjiachun commented 5 years ago

正确的做法是以快照为准，重启时不需要为你自己的状态机额外保存任何状态，如果你的状态机数据做了持久化到磁盘，请启动前删除，依靠 onSnapshotLoad + replay raft log 来恢复，这样数据一致性

另外 onSnapshotLoad 如果返回 false，会将状态机设置为 error 状态并停止状态机

killme2008 commented 5 years ago

@fengjiachun 建议这一点写入到我们的文档

fengjiachun commented 5 years ago

@fengjiachun 建议这一点写入到我们的文档

恩，我来更新一下文档

huzhihong commented 5 years ago

嗯嗯明白了谢谢

fengjiachun commented 5 years ago

不用客气，没有更多问题的话就关闭这个 issue 了

fengjiachun commented 5 years ago

https://github.com/sofastack/sofastack.tech/pull/291

更新了文档中 snapshot 部分