qbatch

Execute shell command lines in parallel on Slurm, S(on) of Grid Engine (SGE), PBS/Torque clusters

qbatch is a tool for executing commands in parallel across a compute cluster. It takes as input a list of commands (shell command lines or executable scripts) in a file or piped to qbatch. The list of commands are divided into arbitrarily sized chunks which are submitted as jobs to the cluster either as individual submissions or an array. Each job runs the commands in its chunk in parallel according to cores. Commands can also be run locally on systems with no cluster capability via gnu-paralel.

qbatch can also be used within python using the qbatch.qbatchParser and qbatch.qbatchDriver functions. qbatchParser will accept a list of command line options identical to the shell interface, parse, and submit jobs. The qbatchDriver interface will accept key-value pairs corresponding to the outputs of the argument parser, and additionally, the task_list option, providing a list of strings of commands to run.

Installation

$ pip install qbatch

Dependencies

qbatch requires python (>2.7) and GNU Parallel. For Torque/PBS and gridengine clusters, qbatch requires the qsub and qstat commands. For Slurm workload manager, qbatch requires the sbatch and squeue commands.

Environment variable defaults

qbatch supports several environment variables to customize defaults for your local system.

$ export QBATCH_PPJ=12                   # requested processors per job
$ export QBATCH_CHUNKSIZE=$QBATCH_PPJ    # commands to run per job
$ export QBATCH_CORES=$QBATCH_PPJ        # commonds to run in parallel per job
$ export QBATCH_NODES=1                  # number of compute nodes to request for the job, typically for MPI jobs
$ export QBATCH_MEM="0"                  # requested memory per job
$ export QBATCH_MEMVARS="mem"            # memory request variable to set
$ export QBATCH_SYSTEM="pbs"             # queuing system to use ("pbs", "sge","slurm", or "local")
$ export QBATCH_NODES=1                  # (PBS-only) nodes to request per job
$ export QBATCH_SGE_PE="smp"             # (SGE-only) parallel environment name
$ export QBATCH_QUEUE="1day"             # Name of submission queue
$ export QBATCH_OPTIONS=""               # Arbitrary cluster options to embed in all jobs
$ export QBATCH_SCRIPT_FOLDER=".qbatch/" # Location to generate jobfiles for submission
$ export QBATCH_SHELL="/bin/sh"          # Shell to use to evaluate jobfile

Command line help

usage: qbatch [-h] [-w WALLTIME] [-c CHUNKSIZE] [-j CORES] [--ppj PPJ]
              [-N JOBNAME] [--mem MEM] [-q QUEUE] [-n] [-v] [--version]
              [--depend DEPEND] [-d WORKDIR] [--logdir LOGDIR] [-o OPTIONS]
              [--header HEADER] [--footer FOOTER] [--nodes NODES]
              [--sge-pe SGE_PE] [--memvars MEMVARS]
              [--pbs-nodes-spec PBS_NODES_SPEC] [-i]
              [-b {pbs,sge,slurm,local,container}] [--env {copied,batch,none}]
              [--shell SHELL]
              ...

Submits a list of commands to a queueing system. The list of commands can be
broken up into 'chunks' when submitted, so that the commands in each chunk run
in parallel (using GNU parallel). The job script(s) generated by qbatch are
stored in the folder .qbatch/

positional arguments:
  command_file          An input file containing a list of shell commands to
                        be submitted, - to read the command list from stdin or
                        -- followed by a single command

optional arguments:
  -h, --help            show this help message and exit
  -w WALLTIME, --walltime WALLTIME
                        Maximum walltime for an array job element or
                        individual job (default: None)
  -c CHUNKSIZE, --chunksize CHUNKSIZE
                        Number of commands from the command list that are
                        wrapped into each job (default: 1)
  -j CORES, --cores CORES
                        Number of commands each job runs in parallel. If the
                        chunk size (-c) is smaller than -j then only chunk
                        size commands will run in parallel. This option can
                        also be expressed as a percentage (e.g. 100%) of the
                        total available cores (default: 1)
  --ppj PPJ             Requested number of processors per job (aka ppn on
                        PBS, slots on SGE, cpus per task on SLURM). Cores can
                        be over subscribed if -j is larger than --ppj (useful
                        to make use of hyper-threading on some systems)
                        (default: 1)
  -N JOBNAME, --jobname JOBNAME
                        Set job name (defaults to name of command file, or
                        STDIN) (default: None)
  --mem MEM             Memory required for each job (e.g. --mem 1G). This
                        value will be set on each variable specified in
                        --memvars. To not set any memory requirement, set this
                        to 0 (default: 0)
  -q QUEUE, --queue QUEUE
                        Name of queue to submit jobs to (defaults to no queue)
                        (default: None)
  -n, --dryrun          Dry run; Create jobfiles but do not submit or run any
                        commands (default: False)
  -v, --verbose         Verbose output (default: False)
  --version             show program's version number and exit

advanced options:
  --depend DEPEND       Wait for successful completion of job(s) with name
                        matching given glob pattern or job id matching given
                        job id(s) before starting (default: None)
  -d WORKDIR, --workdir WORKDIR
                        Job working directory (default:
                        current working directory)
  --logdir LOGDIR       Directory to save store log files (default:
                        {workdir}/logs)
  -o OPTIONS, --options OPTIONS
                        Custom options passed directly to the queuing system
                        (e.g --options "-l vf=8G". This option can be given
                        multiple times (default: [])
  --header HEADER       A line to insert verbatim at the start of the script,
                        and will be run once per job. This option can be given
                        multiple times (default: None)
  --footer FOOTER       A line to insert verbatim at the end of the script,
                        and will be run once per job. This option can be given
                        multiple times (default: None)
  --nodes NODES         (PBS and SLURM only) Nodes to request per job
                        (default: 1)
  --sge-pe SGE_PE       (SGE-only) The parallel environment to use if more
                        than one processor per job is requested (default: smp)
  --memvars MEMVARS     A comma-separated list of variables to set with the
                        memory limit given by the --mem option (e.g.
                        --memvars=h_vmem,vf) (default: mem)
  --pbs-nodes-spec PBS_NODES_SPEC
                        (PBS-only) String to be inserted into nodes= line of
                        job (default: None)
  -i, --individual      Submit individual jobs instead of an array job
                        (default: False)
  -b {pbs,sge,slurm,local,container}, --system {pbs,sge,slurm,local,container}
                        The type of queueing system to use. 'pbs' and 'sge'
                        both make calls to qsub to submit jobs. 'slurm' calls
                        sbatch. 'local' runs the entire command list (without
                        chunking) locally. 'container' creates a joblist and
                        metadata file, to pass commands out of a container to
                        a monitoring process for submission to a batch system.
                        (default: local)
  --env {copied,batch,none}
                        Determines how your environment is propagated when
                        your job runs. "copied" records your environment
                        settings in the job submission script, "batch" uses
                        the cluster's mechanism for propagating your
                        environment, and "none" does not propagate any
                        environment variables. (default: copied)
  --shell SHELL         Shell to use for spawning jobs and launching single
                        commands (default: /bin/sh)

Some examples:

# Submit an array job from a list of commands (one per line)
# Generates a job script in ./.qbatch/ and job logs appear in ./logs/\
# All defaults are inherited from QBATCH_* environment variables
$ qbatch commands.txt

# Submit a single command to the cluster
$ qbatch -- echo hello

# Set the walltime for each job
$ qbatch -w 3:00:00 commands.txt

# Run 24 commands per job
$ qbatch -c24 commands.txt

# Pack 24 commands per job, run 12 in parallel at a time
$ qbatch -c24 -j12 commands.txt

# Start jobs after successful completion of existing jobs with names starting with "stage1_"
$ qbatch --afterok 'stage1_*' commands.txt

# Pipe a list of commands to qbatch
$ parallel echo process.sh {} ::: *.dat | qbatch -

# Run jobs locally with GNU Parallel, 12 commands in parallel
$ qbatch -b local -j12 commands.txt

# Many options don't make sense locally: chunking, individual vs array, nodes,
# ppj, highmem, and afterok are ignored

A python script example:

# Submit jobs to a cluster using the QBATCH_* environment defaults
import qbatch
task_list = ['echo hello', 'echo hello2']
qbatch.qbatchDriver(task_list = task_list)