[ 
https://issues.apache.org/jira/browse/SOLR-12073?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Cao Manh Dat updated SOLR-12073:
--------------------------------
    Description: 
The case here is
 - Leader receive an update and forward the update to replicaA and replicaB
 - replicaA is getting DOWN, hence leader failed to send the update to replicaA
 - in DUP.finish(), leader get current active replicas of it shard, found only 
replicaB
 - the leader does not put replicaA into LIR, and return the update as a 
success. This is not good because replicaA can win the election when it comes 
back, causing consistency problem.

  was:
The case here is
 - Leader receive an update and forward the update to replicaA and replicaB
 - replicaA is getting DOWN, hence leader failed to send the update to replicaA
 - in DUP.finish(), leader get current active replicas of it shard, found only 
replicaB
 ** in 7.2: the leader does not put replicaA into LIR, and return the update as 
a success. This is not good because replicaA can win the election when it comes 
back, causing consistency problem.
 ** in the current code (7.3): the leader does not put replicaA into LIR, but 
return the update as a failure


> Mismatch between getting replicas in DUP.finish() and DUP.setupRequest() can 
> make an update fail
> ------------------------------------------------------------------------------------------------
>
>                 Key: SOLR-12073
>                 URL: https://issues.apache.org/jira/browse/SOLR-12073
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Cao Manh Dat
>            Assignee: Cao Manh Dat
>            Priority: Major
>
> The case here is
>  - Leader receive an update and forward the update to replicaA and replicaB
>  - replicaA is getting DOWN, hence leader failed to send the update to 
> replicaA
>  - in DUP.finish(), leader get current active replicas of it shard, found 
> only replicaB
>  - the leader does not put replicaA into LIR, and return the update as a 
> success. This is not good because replicaA can win the election when it comes 
> back, causing consistency problem.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to