- Packages jobs as PHAR files for speedy and convenient deployment.
- Also generates shell scripts to invoke a job (requires
$HADOOP_HOME
to be defined). - Automatic detection of streaming settings (so it knows about the configured key field separators and lengths et cetera).
- Abstracted input parsing and splitting (only supports line-based input at the moment though).
- Supports custom arguments for the Hadoop invocation.
- Something like
KeyValueTextInputFormat
does not work, partly because passing it as-inputformat
sets the property toStreamingInputFormat
(must investigate why), and-D mapred.input.format.class
seems to have no effect. Unsure how to fix.
- Unit testing capabilities (Mappers and Reducers could be tested locally).
- Support for input formats other than TextInputFormat (only line-based stuff works out of the box right now).
The examples
sub-directory contains a number of examples.
- A functioning Hadoop installation, with namenode, jobtracker and all other components running.
- If you are developing locally, follow the Hadoop Quick Start Guide to set up pseudo-distributed mode.
- The PHAR PHP extension must be enabled, and
phar.readonly
must be set to0
in php.ini if you want to compile jobs.
Create a folder (this folder name will be the job name later on), with a Mapper.php
containing the mapper class, a Reducer.php
containing a reducer class (if desired), and, if you want, an ARGUMENTS
file with additional arguments.
Note: when you have an ARGUMENTS file, you must include full -mapper
and -reducer
commands, see the examples. Any -D
flags in ARGUMENTS must also precede any other switches such as -mapper
or -file
. Every line in ARGUMENTS must be terminated by a backslash because the raw contents of the file are inserted into the generated shell script.
Assuming your job (and thus folder) name is "TpsReportCount", run:
bin/compile.sh TpsReportCount <BUILDDIR>
Note: the build dir must exist and be writeable.
You can pass path names to include in the package and the default timezone to set for scripts to compile.sh
; invoke it without arguments for usage help. The --debug
switch enables debug mode which means the HADOOPHP_DEBUG
constant will be set to true, you can use this to add counter or message emitters to your code that should only run in such a case.
Assuming your job name is "TpsReportCount", run:
path/to/builddir/TpsReportCount.sh <HDFSINPUTPATH...> <HDFSOUTPUTPATH>
Note: you may pass any number of input path names; the last path name given is the output path where results will be written to.
You may also pass the path to a Hadoop config dir (equivalent to the --config
argument of the hadoop
binary):
path/to/builddir/TpsReportCount.sh -c path/to/dir/with/remote-cluster-config <HDFSINPUTPATH...> <HDFSOUTPUTPATH>