You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@solr.apache.org by "Pushkar Raste (Jira)" <ji...@apache.org> on 2021/03/24 20:35:00 UTC
[jira] [Commented] (SOLR-15286) A brand new replica in the legacy
setup reports healthy before replicating index from its leader
[ https://issues.apache.org/jira/browse/SOLR-15286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17308170#comment-17308170 ]
Pushkar Raste commented on SOLR-15286:
--------------------------------------
Here is a solution I am thinking of.
# On startup check if the node is follower.
# If yes, check the index generation on the node and check the replicable index generation on the leader
# Mark the node (follower) health only if its index generation is with a configurable threshold from the leader index generation.
# Once the node has caught with the leader (within the configuration threshold) mark it an instance variable to indicate to indicate a initial sync with leader is done
We can have another option to always ensure that replica is within acceptable index generation threshold from its leader.
> A brand new replica in the legacy setup reports healthy before replicating index from its leader
> ------------------------------------------------------------------------------------------------
>
> Key: SOLR-15286
> URL: https://issues.apache.org/jira/browse/SOLR-15286
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Components: replication (java)
> Affects Versions: 8.8.1
> Reporter: Pushkar Raste
> Priority: Major
>
> In the legacy mode when you add a brand new replica it immediately starts reporting healthy even if it hasn't fully replicated index from its leader. This issue is also exhibited when you shut down a replica and start it after a lot of data is indexed on the leader.
> If the replica reports healthy LB would continue to forward incoming requests to it and this would result in inconsistent data being retuned to the caller.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)