hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chris Douglas (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-3603) Setting spill threshold to 100% fails to detect spill for records
Date Fri, 20 Jun 2008 20:11:45 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-3603?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12606880#action_12606880

Chris Douglas commented on HADOOP-3603:

The findbugs warning is spurious; the additional reference to kvindex in collect() is necessary
because records are not required to call write, neither is the access threatened by contention
because (1) all accesses to kvindex, including accesses in synchronized blocks, are from the
same thread and (2) the circumstance where this may be true- even for a stale value of kvstart-
would have blocked if the record had performed any writes.

The javac warning is due to deprecation of InputFormat::validateInput, implemented in a test

> Setting spill threshold to 100% fails to detect spill for records
> -----------------------------------------------------------------
>                 Key: HADOOP-3603
>                 URL: https://issues.apache.org/jira/browse/HADOOP-3603
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: mapred
>    Affects Versions: 0.17.0
>            Reporter: Chris Douglas
>            Assignee: Chris Douglas
>            Priority: Blocker
>             Fix For: 0.18.0
>         Attachments: 3603-0.patch, 3603-1.patch
> If io.sort.record.percent is set to 1.0, the simultaneous collection and spill is disabled.
However, if one exhausts the offset allocation before the serialization allocation, the limit
will not be detected and the write will block forever.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message