[
https://issues.apache.org/jira/browse/RATIS-229?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16453263#comment-16453263
]
Tsz Wo Nicholas Sze commented on RATIS-229:
-------------------------------------------
> ... by the time the appendEntryRequest times out other requests may have
> been sent.
Good point! We need to update both times then. Otherwise, it won't trigger a
heartbeat.
It is tricky to handle appendEntry timeout since the leader does not know if
the follower is busy working on the appendEntry request, or if it is
unavailable.
> GRpcLogAppender should trigger a heartbeat when appendEntries timeout
> ---------------------------------------------------------------------
>
> Key: RATIS-229
> URL: https://issues.apache.org/jira/browse/RATIS-229
> Project: Ratis
> Issue Type: Bug
> Reporter: Kit Hui
> Assignee: Kit Hui
> Priority: Major
> Attachments: r229_20180420.patch
>
>
> When appendEntries timeout, GRpcLogAppender has already updated
> lastRpcSendTime so that the next heartbeat may not be sent
> immediately.
> This problem is observed in RATIS-227. The killed follower in
> testBasicAppendEntriesWithAllReplication may start a leader election
> after it has been restarted.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
