[
https://issues.apache.org/jira/browse/HAMA-757?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13655844#comment-13655844
]
Edward J. Yoon commented on HAMA-757:
-------------------------------------
We might be able to approximately calculate the size of partition file before
merging.
{code}
SequenceFile.Writer writer = SequenceFile.createWriter(fs, conf,
partitionFile, keyClass, valueClass, CompressionType.NONE);
for (int i = 0; i < files.length; i++) {
LOG.debug("merge '" + files[i].getPath() + "' into " + partitionDir
+ "/" + getPartitionName(partitionID));
{code}
But, I didn't understand your solution exactly yet. Please feel free to upload
your patch. Let's think about it more and discuss.
> The partitioning job output should be un-splitable
> --------------------------------------------------
>
> Key: HAMA-757
> URL: https://issues.apache.org/jira/browse/HAMA-757
> Project: Hama
> Issue Type: Bug
> Components: bsp core
> Affects Versions: 0.6.1
> Reporter: MaoYuan Xian
>
> When the output sequence files from partitioning job are large(bigger than
> two hdfs file block size), the second round of the job (using these sequence
> file as input) will start up more tasks than client want. Some times, this
> uncertainty make the job exceed the cluster slot capacity.
> In the real project, I implemented an new Inputformat which marked as
> un-splitable to solve the problem. Is there any better way?
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
