cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paulo Motta (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-13629) Wait for batchlog replay during bootstrap
Date Thu, 29 Jun 2017 01:22:00 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-13629?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16067550#comment-16067550
] 

Paulo Motta commented on CASSANDRA-13629:
-----------------------------------------

bq. It throws a blocking replay of the batchlog right before starting the native transport
service. I'm not sure about if this is the best place/moment to do so. 

I think the concern on CASSANDRA-13162 was that a bootstrapping node becomes available before
its MV is successfully finished building. So, we should probably only wait for batchlog replay
after bootstrap is finished and [before marking the MV as built|https://github.com/adelapena/cassandra/blob/771c4e1a3762bcc19bdfcd25cb25a01104515a1e/src/java/org/apache/cassandra/service/StorageService.java#L1494].
Given this, the dtests should probably be updated to test the bootstrap scenario.

On normal node starts, we probably shouldn't wait for batchlog to be replayed since the data
was already replicated to the correct number of nodes via the write CL.

bq. Also, the possible exceptions in the initial batchlog replay are passed to the JVM stability
inspector, maybe we should simply stop the JVM in case of a failure in the initial batchlog
replay. 

In the boostrap case we should probably fail the boostrap if there is an error during batchlog
replay.

> Wait for batchlog replay during bootstrap
> -----------------------------------------
>
>                 Key: CASSANDRA-13629
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-13629
>             Project: Cassandra
>          Issue Type: Sub-task
>          Components: Materialized Views
>            Reporter: Andrés de la Peña
>            Assignee: Andrés de la Peña
>
> As part of the problem described in [CASSANDRA-13162|https://issues.apache.org/jira/browse/CASSANDRA-13162],
the bootstrap logic won't wait for the backlogged batchlog to be fully replayed before changing
the new bootstrapping node to "UN" state. We should wait for batchlog replay before making
the node available.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org


Mime
View raw message