[
https://issues.apache.org/jira/browse/PHOENIX-7562?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kadir Ozdemir reassigned PHOENIX-7562:
--------------------------------------
Assignee: Kadir Ozdemir
> Phoenix HA Rearchitecture for Consistent Failover
> -------------------------------------------------
>
> Key: PHOENIX-7562
> URL: https://issues.apache.org/jira/browse/PHOENIX-7562
> Project: Phoenix
> Issue Type: Improvement
> Reporter: Kadir Ozdemir
> Assignee: Kadir Ozdemir
> Priority: Major
>
> Phoenix clusters can be deployed in pairs for improved availability using the
> Dual Cluster Client (Phoenix HA) feature. The current HA architecture
> utilizes two Phoenix clusters and mutations on one cluster are asynchronously
> replicated to the other using HBase asynchronous replication, and accessed
> via the dual cluster client. However, asynchronous replication can result in
> clients temporarily missing recent writes during unplanned failovers,
> violating Phoenix's strong consistency guarantee.
> To address this potential data loss issue, we propose to synchronously log
> mutations to a set of HDFS files in the standby cluster and then apply them
> to Phoenix asynchronously. This synchronous logging, that is, synchronous
> replication of mutations, will be optimized for Phoenix and implemented at
> the Phoenix level.
> Additionally, the Zookeeper dependency will be removed from the Phoenix dual
> cluster client code by moving HA cluster record watching and caching to the
> server side. The main goals of these changes are to provide consistent and
> fast failover in seconds, and a Zookeeper-less client. These goals will
> reshape the Phoenix HA architecture.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
