hadoop-streaming: A simple Hadoop streaming library

[ bsd3, cloud, distributed-computing, library, mapreduce ] [ Propose Tags ]

A simple Hadoop streaming library based on conduit, useful for writing mapper and reducer logic in Haskell and running it on AWS Elastic MapReduce, Azure HDInsight, GCP Dataproc, and so forth.


[Skip to Readme]
Versions [faq] 0.1.0.0, 0.2.0.0, 0.2.0.1, 0.2.0.2, 0.2.0.3
Change log CHANGELOG.md
Dependencies base (>=4.12 && <5), bytestring (==0.10.*), conduit (>=1.3.1 && <1.4), extra (>=1.6.18 && <1.8), text (>=1.2.2.0 && <1.3) [details]
License BSD-3-Clause
Copyright 2020 Ziyang Liu
Author Ziyang Liu <free@cofree.io>
Maintainer Ziyang Liu <free@cofree.io>
Category Cloud, Distributed Computing, MapReduce
Home page https://github.com/zliu41/hadoop-streaming
Bug tracker https://github.com/zliu41/hadoop-streaming/issues
Source repo head: git clone https://github.com/zliu41/hadoop-streaming
Uploaded by zliu41 at 2020-05-18T15:34:43Z
Distributions NixOS:0.2.0.2
Downloads 407 total (158 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Hackage Matrix CI
Docs available [build log]
Last success reported on 2020-05-18 [all 1 reports]

Modules

[Index] [Quick Jump]

Downloads

Maintainer's Corner

For package maintainers and hackage trustees


Readme for hadoop-streaming-0.2.0.3

[back to package description]

A simple Hadoop streaming library based on conduit, useful for writing mapper and reducer logic in Haskell and running it on AWS Elastic MapReduce, Azure HDInsight, GCP Dataproc, and so forth.

Hackage: https://hackage.haskell.org/package/hadoop-streaming

Word Count Example

See the Haddock in HadoopStreaming.Text for a simple word-count example.

A Few Things to Note

ByteString vs Text

The HadoopStreaming module provides the general Mapper and Reducer data types, whose input and output types are abstract. They are usually instantiated with either ByteString or Text. ByteString is more suitable if the input/output needs to be decoded/encoded, for instance using the base64-bytestring library. On the other hand, Text could make more sense if decoding/encoding is not needed, or if the data is not UTF-8 encoded (see below regarding encodings). In general I'd imagine ByteString being used much more often than Text.

The HadoopStreaming.ByteString and HadoopStreaming.Text modules provide some utilities for working with ByteString and Text, respectively.

Encoding

It is highly recommended that your input data be UTF-8 encoded, as this is the default encoding Hadoop uses. If you must use other encodings such as UTF-16, keep in mind the following gotchas:

  • It is not enough that your code can work with the encoding you choose to use:

    • By default, if any of your input files does not end with a UTF-8 representation of newline, i.e., a 0x0A byte, Hadoop streaming will add a 0x0A byte.

    • Likewise, if any line in your mapper output does not contain a UTF-8 representation of tab (0x09), Hadoop streaming will add it at the end of the line.

    This will almost certainly break your job. It may be possible to configure Hadoop streaming and tell it to use other encodings, so that the above behavior is consistent with the encoding you choose to use, but I don't know whether that is the case. I tried -D mapreduce.map.java.opts="-Dfile.encoding=UTF-16BE" but that doesn't seem to work.

  • If you use ByteString as the input type and use Data.ByteString.hGetLine to read lines from the input, be aware that Data.ByteString.hGetLine uses 0x0A bytes as line breaks, so it doesn't work properly for non-UTF-8 encoded input. For example, in UTF-16BE and UTF-16LE, the newline character is encoded as 0x00 0x0A and 0x0A 0x00, respectively.