[
https://issues.apache.org/jira/browse/HADOOP-3149?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12601990#action_12601990
]
Alejandro Abdelnur commented on HADOOP-3149:
--------------------------------------------
For our requirements the MultipleOutputFormat shortcomings (that are handled by
the proposed patch) are:
* It does not support K & V of different types for the different output names.
* A Map/Reduce job cannot use it from both Map and Reduce.
* If used from a Map the job cannot have a Reduce.
* It is an abstract class and it requires different implementations for
different OutputFormats.
>From the M/R developer usage perspective:
* By looking at the M/R code is not obvious that data is not written to a
different output and it falls outside of the M/R I/O flow.
* It is not clear to which output the data is being written unless there is a
clear understanding of the MOF implementation.
IMO, the current MOF has great flexibility but that comes with a complexity
cost in usage and understanding it.
On your commento on not needing the {{InternalFileOutputFormat}}, you are
right, I could make {{MultipleOutputs}} to extends {{FileOutputFormat}} and
have the {{getRecordWriter()}} method there.
On your comment on using the default file system, I did that because the
{{OutputFormat}} interface indicates that is ignored and because the file to be
created goes in the same place of the job output, thus using the job conf for
it.
> supporting multiple outputs for M/R jobs
> ----------------------------------------
>
> Key: HADOOP-3149
> URL: https://issues.apache.org/jira/browse/HADOOP-3149
> Project: Hadoop Core
> Issue Type: New Feature
> Components: mapred
> Environment: all
> Reporter: Alejandro Abdelnur
> Assignee: Alejandro Abdelnur
> Fix For: 0.18.0
>
> Attachments: patch3149.txt, patch3149.txt, patch3149.txt,
> patch3149.txt, patch3149.txt, patch3149.txt, patch3149.txt, patch3149.txt,
> patch3149.txt, patch3149.txt
>
>
> The outputcollector supports writing data to a single output, the 'part'
> files in the output path.
> We found quite common that our M/R jobs have to write data to different
> output. For example when classifying data as NEW, UPDATE, DELETE, NO-CHANGE
> to later do different processing on it.
> Handling the initialization of additional outputs from within the M/R code
> complicates the code and is counter intuitive with the notion of job
> configuration.
> It would be desirable to:
> # Configure the additional outputs in the jobconf, potentially specifying
> different outputformats, key and value classes for each one.
> # Write to the additional outputs in a similar way as data is written to the
> outputcollector.
> # Support the speculative execution semantics for the output files, only
> visible in the final output for promoted tasks.
> To support multiple outputs the following classes would be added to
> mapred/lib:
> * {{MOJobConf}} : extends {{JobConf}} adding methods to define named outputs
> (name, outputformat, key class, value class)
> * {{MOOutputCollector}} : extends {{OutputCollector}} adding a
> {{collect(String outputName, WritableComparable key, Writable value)}} method.
> * {{MOMapper}} and {{MOReducer}}: implement {{Mapper}} and {{Reducer}} adding
> a new {{configure}}, {{map}} and {{reduce}} signature that take the
> corresponding {{MO}} classes and performs the proper initialization.
> The data flow behavior would be: key/values written to the default (unnamed)
> output (using the original OutputCollector {{collect}} signature) take part
> of the shuffle/sort/reduce processing phases. key/values written to a named
> output from within a map don't.
> The named output files would be named using the task type and task ID to
> avoid collision among tasks (i.e. 'new-m-00002' and 'new-r-00001').
> Together with the setInputPathFilter feature introduced by HADOOP-2055 it
> would become very easy to chain jobs working on particular named outputs
> within a single directory.
> We are using heavily this pattern and it greatly simplified our M/R code as
> well as chaining different M/R.
> We wanted to contribute this back to Hadoop as we think is a generic feature
> many could benefit from.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
