[
https://issues.apache.org/jira/browse/SOLR-9913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15888422#comment-15888422
]
Mark Miller commented on SOLR-9913:
-----------------------------------
Seems reasonable to me.
I'd really like to remove the need for this per update fail request. I think
ideally, this request would go through ZK rather than attempting it directly.
The replica would instead just watch the LIR nodes. That is also how I would
like to get rid of the 'leader publishes down for replica issue'. We would not
really want per update updates to ZK though, so we would probably want some
delayed action that collects requests and only talks to ZK once every few
seconds or something.
> LIR should continue on SocketTimeoutException
> ---------------------------------------------
>
> Key: SOLR-9913
> URL: https://issues.apache.org/jira/browse/SOLR-9913
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Cao Manh Dat
> Attachments: SOLR-9913.patch
>
>
> When I run jepsen tests on latest source. Some node can not recovery on time
> because LIR did not continue trying on SocketTimeoutException.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
