[jira] [Updated] (TEZ-2604) PipelinedSorter doesn't use number of items when creating SortSpan

["Tsuyoshi Ozawa (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/TEZ-2604?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Tsuyoshi Ozawa updated TEZ-2604:
--------------------------------
    Attachment: TEZ-2604.001.patch

Attaching a first patch.

> PipelinedSorter doesn't use number of items when creating SortSpan 
> -------------------------------------------------------------------
>
>                 Key: TEZ-2604
>                 URL: https://issues.apache.org/jira/browse/TEZ-2604
>             Project: Apache Tez
>          Issue Type: Bug
>    Affects Versions: 0.8.0
>            Reporter: Tsuyoshi Ozawa
>            Assignee: Tsuyoshi Ozawa
>         Attachments: TEZ-2604.001.patch
>
>
> {quote}
>       int items = 1024*1024;
>       int perItem = 16;
>       if(span.length() != 0) {
>         items = span.length();
>         perItem = span.kvbuffer.limit()/items;
>         items = (int) ((span.capacity)/(METASIZE+perItem));
>         if(items > 1024*1024) {
>             // our goal is to have 1M splits and sort early
>             items = 1024*1024;
>         }
>       }
>       Preconditions.checkArgument(listIterator.hasNext(), "block iterator should not be empty");
>       span = new SortSpan((ByteBuffer)listIterator.next().clear(), (1024*1024),
>           perItem, ConfigUtils.getIntermediateOutputKeyComparator(this.conf));
> {quote}
> Should we use items instead of (1024*1024)?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)