In caravagnalab/easypar: Easy parallaleization of R functions.

easypar can be used to generate scripts that submit array jobs to the PBSpro cluster system.

First, write in R your code.

# Computation function 
FUN = function(x, y){  ... }

# Input for 25 array jobs that match FUN arguments
PARAMS = data.frame(x = runif(25), y = runif(25))

# Generates submission files 
run_PBSpro(FUN, PARAMS)

Then, in your terminal.

# Test if the generated script runs with the first input 
head -1 EASYPAR_PBSpro_input_jobarray.csv | Rscript EASYPAR_PBSpro_Run.R

# Submit array jobs (after loading the cluster module)
qsub < EASYPAR_PBSpro_submission.sh

Requirements

a function FUN (e.g.,: FUN = ls) that can run as a stand-alone R application, with its own parameters and that manages explicitly its dependencies.

Note: FUN will be run as an indipendent process.

a dataframe PARAMS where every row is an input for FUN. The column order must match FUN arguments.

Conceptually, you set up the data as for an apply(FUN, MARGIN = 1) by row.

The input should have column names without dots or spaces; these will match the arguments of FUN. So, for instance, an input with 2-columns will only work if FUN has 2 parameters.

Output

run_PBSpro generates 3 files:

an R script wrapping the definition of FUN, with extra code to call FUN using parameters from the command line. Your function in this script is called with a fake name;
a csv file containing the input PARAMS, without any header (column names), and row names.
a PBSpro array job submission script with N jobs where N are the rows of PARAMS.

Before submitting the job, test the computation as explained above.

Customising jobs

Cluster-specific QSUB instructions can be specified, as well as other dependencies from modules available on the cluster.

Function run_PBSpro allows to:

specify a list of modules that will be added as dependencies of the PBSpro submission script. For instance, modules = 'R/3.5.0' will generate the dependecy for a specific R version (3.5.0).
customize the QSUB parameters of the generated script.

The package comes with a default QSUB configuration, that has to be updated according to your cluster setup.

library(easypar)

# Default parameters in the package
default_QSUB_config()

These are classical QSUB parameters: -P = the project ID, -q = the queue ID, -l walltime = the wall time of the jobs, -l nodes=:ppn= = the number of nodes and cpus to allocate as resources, -N = the job ID, -o and -e = the output and error filenames. Notice that by default we have the job array ID in the filename, so to have one log per job.

It is required to modify the default values of -P and -q, the project and queue ID, according to your PBSpro configuration. Otherwise, the submission script will generate an error becaue the default values do not mean anything.

Modifications are done to the default list of parameters; other QSUB flags can be used as well. No checkings on their correctness are done by easypar.

custom_QSUB = default_QSUB_config()

# More informative job ID
custom_QSUB$`-J` = "bwa_aligner"

# A token for a project allowed to run on the cluster
custom_QSUB$`-P` = "DKSMWOP331"

# A queue name that is valid on the cluster
custom_QSUB$`-q` = "bioinformatics"

print(custom_QSUB)

# Shorter version
custom_QSUB = default_QSUB_config(J = 'bwa_aligner', project = 'DKSMWOP331', queue = 'bioinformatics')

print(custom_QSUB)

Once the QSUB has been customized, you can either:

generate the submission scripts, and submit your job manually
generate the submission scripts and submit the jobs with a system call.

By default (run = FALSE) the run_PBSpro function outputs the shell command that should be used to submit the jobs, but leaves the user to submit the job. This is because we experienced some command line issues calling modules with a system call.

Example

An example computation follows.

# A simple function that prints some outputs
FUN = function(x, y){ 
  print(x, y) 
}

# input for 25 array jobs
PARAMS = data.frame(
  x = runif(25), 
  y = runif(25)
  )

# generates the input files, adding some module dependencies
run_PBSpro(FUN, 
        PARAMS, 
        QSUB_config = custom_QSUB,
        modules = 'R/3.5.0'
        )