A simple script for submitting

gistfile1.txt

# farm_scripts.R - a simple script to submit and run LSF jobs and job
#   arrays from R. The "main" function reads in arguments from the
#   command-line and passes them, possibly along with the job array
#   index, to a user-defined function. This can serve as a basic tool
#   for creating and running batched processes using R in an LSF 
#   environment.
#
#  contact: greg@ebi.ac.uk

# Extract the command-line arguments
args <- commandArgs(trailingOnly=T)

main <- function() {
  if (!is.na(args[1])) {
    # The first arg is the function name
    fn.name <- args[1]
    print(paste("Calling", fn.name))

    # Get the LSB_JOBINDEX environment variable.
    # If it exists, this indicates that the script is being run
    # as part of a job array and gives us the array index.
    # Usually this is used to determine which slice of data
    # the function should process.
    jobindex <- Sys.getenv('LSB_JOBINDEX')
    if (jobindex != '' && as.integer(jobindex) > 0) {
      args <- c(args, as.integer(jobindex))
    }
    print(args)

    # Print out some info if an error occurs.
    error.f <- function(e) {
      print("###ERROR###")
      print(as.character(e))
    }

    # Run the indicated function, using the command-line arguments as parameters.
    tryCatch(
      do.call(fn.name, as.list(args[2:length(args)])),
      error = error.f
    )
  }
}

# A simple test function to verify that parameters are being passed
# correctly. In a real-world scenario, your main analysis code would go here.
test.function <- function(argA='', argB='', jobindex=0) {
  print(paste("Running job", jobindex, "with argA=", argA, "and argB=", argB))
}

# Submits a job to the LSF environment. This job will open an R process and use
# this same script as input. The name of the desired function (e.g., "test.function")
# will be passed to the submitted job, along with any additional parameters,
# as command-line arguments. Job arrays are also supported.

bsub.function <- function(fn_name, queue='normal', mem=4, args='', jobarray.count=NULL, jobarray.id=fn_name) {
  array_s = ''
  if (!is.null(jobarray.count)) {
    # Optionally create a job array
    array_s <- paste('-J ', jobarray.id, '[1-', jobarray.count, ']', sep='')
  }
  
  args_s <- paste(fn_name, ' ', args, sep='')
  uname  <- Sys.getenv("USER")

  # Note that the job output logs will be stored in the home directory. If you are running a large
  # job array, this will quickly clutter up the directory with files. If several thousand jobs are
  # run, this will overload the filesystem (BAD). You could send output to /dev/null to avoid this.
  cmd <- sprintf('bsub -q research -M%d000 -R "rusage[mem=%d000]" %s -o "/homes/%s/lsf_log_%s_%%J_%%I.txt" "R-2.12.0 --vanilla --args %s < farm_scripts.R"',
      mem, mem, array_s, uname, fn_name, args_s
  )
  print(cmd)
  system(cmd)
}

# Run the "main" function only after all other functions have been defined.
main()