[
https://issues.apache.org/jira/browse/HDDS-1621?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16855621#comment-16855621
]
Supratim Deka commented on HDDS-1621:
-------------------------------------
keeping the FileChannel around after writeData and passing it back to the State
Machine is not really required.
dfs.container.chunk.write.sync determines whether the chunk is persisted as
soon as the data is written.
If this parameter is set to false, it implies the possibility of data loss.
This tradeoff is provided to enable higher throughput.
In keeping with this understanding, we will limit the change to
1. invoking a force+close on the channel inside writeData if the sync option is
set.
2. change AsynchronousFileChannel to FileChannel (as explained in the previous
comment)
> flushStateMachineData should ensure the write chunks are flushed to disk
> ------------------------------------------------------------------------
>
> Key: HDDS-1621
> URL: https://issues.apache.org/jira/browse/HDDS-1621
> Project: Hadoop Distributed Data Store
> Issue Type: Sub-task
> Reporter: Shashikant Banerjee
> Assignee: Supratim Deka
> Priority: Major
>
> Currently, chunks writes are not synced to disk by default. When
> flushStateMachineData gests invoked from Ratis, it should also ensure all the
> pending chunk writes should be flushed to disk.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
