In order to prevent data loss, new servers should be formatted explicitly. Auto-format may cause data loss as described in https://issues.apache.org/jira/browse/RATIS-1677 .
Tsz-Wo On Tue, Jan 9, 2024 at 5:32 PM tison <[email protected]> wrote: > Thank you! Now I can make things work with > `.setOption(RaftStorage.StartupOption.RECOVER)` on > `RaftServer.newBuilder()`. > > I noticed that [1] enables fallback to format if recovery fails. And > it meets my intuition that we can always try to recover and initialize > if it fails. Still, the initialization can fail with "Failed to > FORMAT: One or more existing directories found". I wonder when it's > desired to configure `.setOption(RaftStorage.StartupOption.FORMAT)` > since [1] is there. > > Best, > tison. > > [1] https://github.com/apache/ratis/pull/903 > > Tsz Wo Sze <[email protected]> 于2024年1月10日周三 08:57写道： > > > > > ... between the two starting-ups. > > > > It is related to RaftStorage.StartupOption -- the default is FORMAT so > the first time can successfully start up. However, it failed to start up > the second time since the directory was already formatted. We should use > the RECOVER option for the subsequent startups. This change was for > https://issues.apache.org/jira/browse/RATIS-1677 . > > > > Sorry for the inconvenience. > > > > Tsz-Wo > > > > > > On Tue, Jan 9, 2024 at 4:40 PM tison <[email protected]> wrote: > >> > >> Hi, > >> > >> It's a bit unexpected that startup and shutdown a ratis server with an > >> empty state machine (extends BaseStateMachine without any override) > >> returns this error: > >> > >> > Failed to FORMAT: One or more existing directories found > >> > >> The data under the directory should be compatible between the two > starting-ups. > >> > >> You can reproduce it with [1] running EtcdServer's main function twice. > >> > >> I don't know why I encountered this exception and if it's a bug or if > >> I should modify the application logic. > >> > >> Best, > >> tison. > >> > >> [1] https://github.com/tisonkun/ratis-etcd >
