[jira] [Comment Edited] (RATIS-291) Raft Server should fail themselves when a raft storage directory fails

["Shashikant Banerjee (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/RATIS-291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16585672#comment-16585672 ] 

Shashikant Banerjee edited comment on RATIS-291 at 8/20/18 9:44 AM:
--------------------------------------------------------------------

In Patch v1 , the leader steps down in case the raft log worker encounters an error while writing/truncating the log file or in case of any stateMachineException thrown while applying the log.

 

Statemachine API to verify the log location can be addded as a separate Jira.


was (Author: shashikant):
In Patch v1 , the leader steps down in case the raft log worker encounters an error while writing/truncating the log file or in case of any stateMachineException thrown while applying the log.

> Raft Server should fail themselves when a raft storage directory fails
> ----------------------------------------------------------------------
>
>                 Key: RATIS-291
>                 URL: https://issues.apache.org/jira/browse/RATIS-291
>             Project: Ratis
>          Issue Type: Bug
>          Components: server
>    Affects Versions: 0.3.0
>            Reporter: Mukul Kumar Singh
>            Assignee: Shashikant Banerjee
>            Priority: Major
>              Labels: ozone
>             Fix For: 0.3.0
>
>         Attachments: RATIS-291.01.patch
>
>
> A Raft server uses a storage directory to store the write ahead log. If this log is lost because of a reason, then this node should fail itself.
> For a follower, if raft log location has failed, then the follower will not be able to append any entries. This node will now be lagging behind the follower and will eventually be notified via notifySlowness.
> For a leader where the raft log disk has failed, the leader will not append any new entries to its log. However with respect to the raft ring, the leader will still remain healthy. This jira proposes to add a new api to identify a leader with failed node.
> Also this jira also proposes to add a new api to the statemachine, so that state machine implementation can provide methods to verify the raft log location.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)