condor

The goal of condor is to run R batch commands in HTCondor on a remote cluster directly from a local terminal or RStudio work environment.

Installation

You can install condor from github with:

# install.packages("devtools")
devtools::install_github("yuliasidi/condor")

SSH Key Setup

• Installation of the R package ssh
• setting up an ssh key
  • create a key pairing on the local machine
  • on the remote machine
    • In the user directory (~) create the subdirectory ~/.ssh if it is not already there
      • mkdir ~/.ssh
    • Create a file in ~/.ssh called authorized_keys
      • touch ~/.ssh/authorized_keys
  • copy the public key contents to the remote server into .ssh/authorized_keys
    • cat ~/.ssh/id_rsa.pub | pbcopy (mac)
  • paste the contents of the clipboard on the remote into ~/.ssh/authorized_keys
    • echo '[PASTE CONTENTS OF CLIPBOARD]' > ~/.ssh/authorized_keys

Workflow

Load Libraries

library(condor)
library(ssh)

Preprocessing

Populate package template and create Rcalcpi.condor in example subdir.

condor::build_template(
  file = 'calcpi.R',
  args = c('$(Process)'),
  tag = 'pi',
  jobs = 5,
  init_dir = 'jobs/run',
  template_file = 'example/Rcalcpi.condor',
  job_type = 'test')

Lines in file that will be run on the cluster

readLines(system.file('example/calcpi.R',package='condor'))
#>  [1] "#!/home/statsadmin/R/bin/Rscript"          
#>  [2] ""                                          
#>  [3] "# Prepare: collect command line arguments,"
#>  [4] "# set iteration number and a unique seed"  
#>  [5] "args <- commandArgs()"                     
#>  [6] "set.seed(Sys.time())"                      
#>  [7] "n <- as.numeric(args[length(args)-1])"     
#>  [8] ""                                          
#>  [9] "# Collect n samples"                       
#> [10] "x <- runif(n,k)"                           
#> [11] "y <- runif(n)"                             
#> [12] ""                                          
#> [13] "# Compute and output the value of pi"      
#> [14] "pihat <- sum(x * x + y * y < 1) / n * 4"   
#> [15] "pihat"                                     
#> [16] "write(pihat, args[length(args)])"          
#> [17] "proc.time()"

Lines in the populated condor file

readLines(system.file('example/Rcalcpi.condor',package='condor'))
#>  [1] "executable = calcpi.R"                                    
#>  [2] "universe = vanilla"                                       
#>  [3] "Requirements = ParallelSchedulingGroup == \"stats group\""
#>  [4] "+AccountingGroup = \"group_statistics_testjob.yuliasidi\""
#>  [5] ""                                                         
#>  [6] "should_transfer_files = YES"                              
#>  [7] "when_to_transfer_output = ON_EXIT"                        
#>  [8] ""                                                         
#>  [9] "arguments = $(Process)"                                   
#> [10] "output    = out/pi-$(Process).Rout"                       
#> [11] "error     = err/pi-$(Process).err"                        
#> [12] "log       = log/pi.log"                                   
#> [13] ""                                                         
#> [14] "initialdir = jobs/run"                                    
#> [15] "# transfer_input_files ="                                 
#> [16] "# transfer_output_files ="                                
#> [17] ""                                                         
#> [18] "Queue 5"

Connect to ssh

session <- ssh::ssh_connect(Sys.getenv('UCONN_USER'))

Upload files needed for the job to the cluster

ssh::scp_upload(session,
                files = c('example/calcpi.R',
                          'example/Rcalcpi.condor', 
                          'example/emailMyself.txt'),
                to = '~'
)

Create directories needed for job outputs

condor::create_dirs(session, file = 'example/Rcalcpi.condor')

Submit the jobs

condor::condor_submit(session,'Rcalcpi.condor')

During the job

condor::condor_q(session)

condor::condor_rm(session,'5000.1')

Post Processing

Retrieve the files

condor::pull(session,
             from = c('jobs/run/log',
                      'jobs/run/out',
                      'jobs/run/err',
                      'jobs/run/*.rds'),
             to = c('output',
                    'output',
                    'output',
                    'output/data'))

Remove files from the cluster

condor::cleanup_remote(session)

Collect output from the local into a single object

RETURN <- purrr::map(
  list.files(path = "example/output/data",full.names = TRUE),
  readRDS
  )

Remove files from the local machine

condor::cleanup_local(dir = 'example/output',tag = 'pi')

Close ssh connection

ssh::ssh_disconnect(session)