[
https://issues.apache.org/jira/browse/RATIS-624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17185513#comment-17185513
]
Rui Wang commented on RATIS-624:
--------------------------------
Thanks [~szetszwo]. This is exactly the discussion I was looking for.
{quote}1. add pause() and resume() to RaftServerImpl so that they can be used
internally{quote}
It makes sense to start from having pause() and resume() in RaftServerImpl. I
think whether needs such API in client or not is decided by which side will
perform pause/install snapshot/resume operations. If it will happen within
Ratis we might not need to expose it in client.
{quote}2. change server detect itself for lagging behind by a configurable
threshold.{quote}
Either Ratis or applications can do this job. But Ratis definitely fits better
here cause it has all stats already. so +1 to leave this is a a configurable
threshold and servers can detect themselves for lagging too far.
{quote}3. change server to ask for installing snapshot if it detects itself
lagging behind. This task need a more detailed design{quote}
I don't have clear thought right now. Either leaders or applications can make
decisions to start pause->snapshot->install snapshot->resume. But as you have
pointed out, taking snapshot might have different cost for applications, so
applications might require different polity for this process.
So my current thought is, convert 1 and 2 to two JIRAs. Then based on
implementation of 1 and 2, combined with the requirement from OZone to have a
design for 3.
What do you think?
> RaftServer should support pause/ unpause in its LifeCycle state
> ---------------------------------------------------------------
>
> Key: RATIS-624
> URL: https://issues.apache.org/jira/browse/RATIS-624
> Project: Ratis
> Issue Type: Task
> Reporter: Hanisha Koneru
> Assignee: Rui Wang
> Priority: Major
> Labels: ozone
> Fix For: 1.1.0
>
> Time Spent: 40m
> Remaining Estimate: 0h
>
> This Jira aims to add support to RaftServer to support pause and unpause to
> its state. When paused, the RaftServer should not accept any incoming append
> log entries.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
