[
https://issues.apache.org/jira/browse/CASSANDRA-7776?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Paul Pak updated CASSANDRA-7776:
--------------------------------
Description:
After sstable files are written, all files in the specified output directory
are loaded (transferred) to the remote cassandra cluster. If multiple writes
occur on a node to the same table (i.e. directory), then the multiple load
processes end up transferring the same sstable files multiple times.
Furthermore, if directory cleanup of successful outputs is set to occur
([CASSANDRA-7777|https://issues.apache.org/jira/browse/CASSANDRA-7777]), then
there could be errors caused by write/load contention.
This can be simply remedied by using unique output directories for each MR job.
was:
After sstable files are written, all files in the specified output directory
are loaded (transferred) to the remote cassandra cluster. If multiple writes
occur on a node to the same table (i.e. directory), then the multiple load
processes end up transferring the same sstable files multiple times.
Furthermore, if directory cleanup of successful outputs is set to occur
([CASSANDRA-7777|]), then there could be errors caused by write/load contention.
This can be simply remedied by using unique output directories for each MR job.
> Allow multiple MR jobs to concurrently write to the same column family from
> the same node using CqlBulkOutputFormat
> -------------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-7776
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7776
> Project: Cassandra
> Issue Type: Improvement
> Components: Hadoop
> Reporter: Paul Pak
> Assignee: Paul Pak
> Priority: Minor
> Labels: cql3, hadoop
> Attachments: trunk-7776-v1.txt
>
>
> After sstable files are written, all files in the specified output directory
> are loaded (transferred) to the remote cassandra cluster. If multiple writes
> occur on a node to the same table (i.e. directory), then the multiple load
> processes end up transferring the same sstable files multiple times.
> Furthermore, if directory cleanup of successful outputs is set to occur
> ([CASSANDRA-7777|https://issues.apache.org/jira/browse/CASSANDRA-7777]), then
> there could be errors caused by write/load contention.
> This can be simply remedied by using unique output directories for each MR
> job.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
