Spark-pipe is a tool for spark, that changes a command line program(C/C++, Java, etc.) of stand-alone version to a distributed one. Spark-pipe makes use of pipe-API to implement the computing task by preprocessing(merging) the source data and spliting it before executing spark map job. Its process flow could be showed as follows.

• clone the repo
• cd to the project directory

$ make

After compiling, there would create a build folder and get logs like this :

================================================================================
The Program has been compiled.
Location    = /xxxx/spark-pipe/build/
JAVA_HOME   = /xxxx/jdk1.8.0_77
HADOOP_HOME = /usr/local/hadoop
SPARK_HOME = /usr/local/spark
================================================================================

Step 1 :Modify these script according your program

• merge.sh It's a executable program you change one data unit to one string line. For example :

example_1.fastq :

@ERR000589.36 EAS139_45:5:1:2:1002/1
AATTGTCCAAGAGATTCTCACACATGTGAAAAATGACTGGTATAGAAGATT
+
IAIIIIIIIIIII.6ID5ID=<>>II+I2A.H.I;I.<09+2&06+&+)/&

example_2.fastq :

@ERR000589.36 EAS139_45:5:1:2:1002/2
TCTTCACCATTCTTTTATAGCAATACATAGTATTCCATCTTAGGCCTGTTC
+
3IIFHIIGFI1IIII<I5<7,=@F4332/D+13C58:/*//'?/')*:27.

merge.fastq

@ERR000589.36 EAS139_45:5:1:2:1002/1|AATTGTCCAAGAGATTCTCACACATGTGAAAAATGACTGGTATAGAAGATT|+|IAIIIIIIIIIII.6ID5ID=<>>II+I2A.H.I;I.<09+2&06+&+)/&|@ERR000589.36 EAS139_45:5:1:2:1002/2|TCTTCACCATTCTTTTATAGCAATACATAGTATTCCATCTTAGGCCTGTTC|+|3IIFHIIGFI1IIII<I5<7,=@F4332/D+13C58:/*//'?/')*:27.

• split_program.py Split one line to several lines if necessary.

• extern_program.sh
  A command line program implemented using C/C++, Java, etc.

Step 2 :Run bash script

USAGE: ./build/pipe_execute.sh SPARK_MASTER HDFS_FILE HDFS_FOLDER PARTITION_NUM

Take the data & conf of current repo for example :

$ cd data
$ ../conf/merge.sh example_1.fastq example_2.fastq merge.fastq
$ hdfs dfs -copyFromLocal merge.fastq $HDFSPATH/merge.fastq
$ cd ../bin
$ ./pipe_execute.sh $HDFSPATH/merge.fastq $HDFSPATH/OUTPUT 1

• Make sure to DOWLOAD the extern programe writen in :

${ROOT_DIR}/conf/extern_program.sh

e.g.

/xxx/bwa mem -t 16 /xxx/hg19Index/hg19.fa ${LINE}

1. ${LINE} actually is like "1.fastq 2.fastq". See output format of split_program.py.
2. bwa download
3. hg19Index download

• Make sure your conf path and extern_program can be accessed by all your computer cluster nodes if using spark cluster mode.

Apache 2.0