[
https://issues.apache.org/jira/browse/HDDS-3306?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Arpit Agarwal updated HDDS-3306:
--------------------------------
Target Version/s: 0.7.0
> Support configurable data consistency level
> -------------------------------------------
>
> Key: HDDS-3306
> URL: https://issues.apache.org/jira/browse/HDDS-3306
> Project: Hadoop Distributed Data Store
> Issue Type: Improvement
> Reporter: Sammi Chen
> Priority: Major
>
> Currently，Ozone provides strong data consistency. On data write, Ozone client
> uses RaftProtos.ReplicationLevel.ALL_COMMITTED which guarantees that the data
> is persisted to all three datanodes disk before ack the write finish. On data
> read, if topology is enabled, client will choose the closest datanode to read
> from. If topology is disabed, then client will always read from leader
> datanode.
> In some cases, user can tolerate weaker data consistency. They want to
> balance data consistency with read/write throughput, just as some of our
> customer required.
> So this issue is to propose a client side configurable consistency level,
> something like Apache cassandra provides, but not that complicated. The
> consistency level will be a write and read combination, for example，
> 1. strong consistency, write: ALL, read:any
> 2. strong consistency-favor write throughput, write: majority, read:leader
> 3. evental consistency, write: majority, read: any
> 4. strong consistency-no data safety guarantee，write:leader, read: leader
> 5. evental consistency-no data safety guarantee，write: leader, read:any
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
