[
https://issues.apache.org/jira/browse/SOLR-6237?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mark Miller updated SOLR-6237:
------------------------------
Attachment: 0001-unified.patch
Here is a patch demonstrating the approach I'm taking.
It's a bit stale compared to my current working state, but in trying to fix the
leader promotion phase, I've walked back a few things as much as I've walked
others forward.
This shows the basic idea though.
* Only the leader writes to the index or tlog, the replicas are in read only
mode.
* Data dirs are published in ZK so that replicas can use the same location as
leaders - this requires some juggling of init logic on startup.
* We have to be careful not to use a Writer when reopening a Reader - hard
commits will control visibility in this mode as well, not soft commits.
* Realtime get is forwarded from replicas to the leader - non leaders can't
serve them.
* Split shard and migrate and the like may need some additional work to operate
correct.
* There are still issue around index locking I need to clear up.
* There are still some problems I'm working through on leader promotion
involving seeing a corrupt view of the index.
* More tests are needed.
> An option to have only leaders write and replicas read when using a shared
> file system with SolrCloud.
> ------------------------------------------------------------------------------------------------------
>
> Key: SOLR-6237
> URL: https://issues.apache.org/jira/browse/SOLR-6237
> Project: Solr
> Issue Type: New Feature
> Components: hdfs, SolrCloud
> Reporter: Mark Miller
> Assignee: Mark Miller
> Attachments: 0001-unified.patch, Unified Replication Design.pdf
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
