[
https://issues.apache.org/jira/browse/SOLR-9223?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Noble Paul updated SOLR-9223:
-----------------------------
Description:
We ended up with a leaderless shard in the following scenario
1) Add a new replica to a shard
2) while the node is still recovering do indexing
3) Delete the replica that is the leader of the shard
It's likely that the latest update(s) is only sent to the node that is
recovering. Other nodes try to become the leader but they can't because they
don't have the latest updates. The node that has the latest updates is trying
to recover. And we end up in a situation where the shard does not have a leader.
How do we resolve this problem? Any suggestions [~markrmil...@gmail.com]
was:
We ended up with a leaderless shard in the following scenario
1) Add a new replica to a shard
2) while the node is still recovering do indexing
3) Delete the replica that is the leader of the shard
It's likely that the latest update(s) is only sent to the node that is
recovering. Other nodes try to become the leader but they can't because they
don't have the latest updates. The node that has the latest updates is trying
to recover. And we end up in a situation where the shard does not have a leader.
This problem is introduced since SOLR-3813 because now we consult even
non-ACTIVE nodes also for recovery
How do we resolve this problem? Any suggestions [~markrmil...@gmail.com]
> Unable to create a shard leader if only a recovering node has the latest
> updates
> --------------------------------------------------------------------------------
>
> Key: SOLR-9223
> URL: https://issues.apache.org/jira/browse/SOLR-9223
> Project: Solr
> Issue Type: Bug
> Reporter: Noble Paul
> Assignee: Noble Paul
>
> We ended up with a leaderless shard in the following scenario
> 1) Add a new replica to a shard
> 2) while the node is still recovering do indexing
> 3) Delete the replica that is the leader of the shard
> It's likely that the latest update(s) is only sent to the node that is
> recovering. Other nodes try to become the leader but they can't because they
> don't have the latest updates. The node that has the latest updates is trying
> to recover. And we end up in a situation where the shard does not have a
> leader.
> How do we resolve this problem? Any suggestions [~markrmil...@gmail.com]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
