User Guide

Setting up the scheduler

To set up a scheduler explicitly, see the following links:

• SLURM - should work without setup
• LSF - should work without setup
• SGE - may require configuration
• PBS/Torque - needs options(clustermq.scheduler="PBS"/"Torque")
• you can suggest another scheduler by opening an issue

You may in addition need to activate compute environments or containers if your shell (e.g. ~/.bashrc) does not do this automatically.

Check the FAQ if your job submission/call to Q errors or gets stuck.

Local parallelization

While this is not the main focus of the package, you can use it to parallelize function calls locally on multiple cores or processes. This can also be useful to test your code on a subset of the data before submitting it to a scheduler.

• Multiprocess (recommended) - Use the callr package to run and manage multiple parallel R processes with options(clustermq.scheduler="multiprocess")
• Multicore - Uses the parallel package to fork the current R process into multiple threads with options(clustermq.scheduler="multicore"). This sometimes causes problems (macOS, RStudio) and is not available on Windows.

SSH connector

There are reasons why you might prefer to not to work on the computing cluster directly but rather on your local machine instead. RStudio is an excellent local IDE, it’s more responsive than and feature-rich than browser-based solutions (RStudio server, Project Jupyter), and it avoids X forwarding issues when you want to look at plots you just made.

Using this setup, however, you lost access to the computing cluster. Instead, you had to copy your data there, and then submit individual scripts as jobs, aggregating the data in the end again. clustermq is trying to solve this by providing a transparent SSH interface.

In order to use clustermq from your local machine, the package needs to be installed on both there and on the computing cluster. On the computing cluster, set up your scheduler and make sure clustermq runs there without problems. Note that the remote scheduler can not be LOCAL (default if no HPC scheduler found) or SSH for this to work.

# If this is set to 'LOCAL' or 'SSH' you will get the following error:
#  Expected PROXY_READY, received ‘PROXY_ERROR: Remote SSH QSys is not allowed’
options(
    clustermq.scheduler = "multiprocess" # or multicore, LSF, SGE, Slurm etc.
)

On your local machine, add the following options:

options(
    clustermq.scheduler = "ssh",
    clustermq.ssh.host = "user@host", # use your user and host, obviously
    clustermq.ssh.log = "~/cmq_ssh.log" # log for easier debugging
)

We recommend that you set up SSH keys for password-less login.

The Q function

The following arguments are supported by Q:

• fun - The function to call. This needs to be self-sufficient (because it will not have access to the master environment)
• ... - All iterated arguments passed to the function. If there is more than one, all of them need to be named
• const - A named list of non-iterated arguments passed to fun
• export - A named list of objects to export to the worker environment

Behavior can further be fine-tuned using the options below:

• fail_on_error - Whether to stop if one of the calls returns an error
• seed - A common seed that is combined with job number for reproducible results
• memory - Amount of memory to request for the job (bsub -M)
• n_jobs - Number of jobs to submit for all the function calls
• job_size - Number of function calls per job. If used in combination with n_jobs the latter will be overall limit
• chunk_size - How many calls a worker should process before reporting back to the master. Default: every worker will report back 100 times total

The full documentation is available by typing ?Q.

Examples

The package is designed to distribute arbitrary function calls on HPC worker nodes. There are, however, a couple of caveats to observe as the R session running on a worker does not share your local memory.

The simplest example is to a function call that is completely self-sufficient, and there is one argument (x) that we iterate through:

fx = function(x) x * 2
Q(fx, x=1:3, n_jobs=1)
#> Running sequentially ('LOCAL') ...
#> [[1]]
#> [1] 2
#> 
#> [[2]]
#> [1] 4
#> 
#> [[3]]
#> [1] 6

Non-iterated arguments are supported by the const argument:

fx = function(x, y) x * 2 + y
Q(fx, x=1:3, const=list(y=10), n_jobs=1)
#> Running sequentially ('LOCAL') ...
#> [[1]]
#> [1] 12
#> 
#> [[2]]
#> [1] 14
#> 
#> [[3]]
#> [1] 16

If a function relies on objects in its environment that are not passed as arguments (including other functions), they can be exported using the export argument:

fx = function(x) x * 2 + y
Q(fx, x=1:3, export=list(y=10), n_jobs=1)
#> Running sequentially ('LOCAL') ...
#> [[1]]
#> [1] 12
#> 
#> [[2]]
#> [1] 14
#> 
#> [[3]]
#> [1] 16

If we want to use a package function we need to load it on the worker using the pkgs parameter, or referencing it with package_name:::

f1 = function(x) splitIndices(x, 3)
Q(f1, x=3, n_jobs=1, pkgs="parallel")
#> Running sequentially ('LOCAL') ...
#> [[1]]
#> [[1]][[1]]
#> [1] 1
#> 
#> [[1]][[2]]
#> [1] 2
#> 
#> [[1]][[3]]
#> [1] 3

f2 = function(x) parallel::splitIndices(x, 3)
Q(f2, x=8, n_jobs=1)
#> Running sequentially ('LOCAL') ...
#> [[1]]
#> [[1]][[1]]
#> [1] 1 2 3
#> 
#> [[1]][[2]]
#> [1] 4 5
#> 
#> [[1]][[3]]
#> [1] 6 7 8

# Q(f1, x=5, n_jobs=1)
# (Error #1) could not find function "splitIndices"

As parallel foreach backend

The foreach package provides an interface to perform repeated tasks on different backends. While it can perform the function of simple loops using %do%:

library(foreach)
foreach(i=1:3) %do% sqrt(i)
#> [[1]]
#> [1] 1
#> 
#> [[2]]
#> [1] 1.414214
#> 
#> [[3]]
#> [1] 1.732051

it can also perform these operations in parallel using %dopar%:

foreach(i=1:3) %dopar% sqrt(i)
#> Warning: executing %dopar% sequentially: no parallel backend registered
#> [[1]]
#> [1] 1
#> 
#> [[2]]
#> [1] 1.414214
#> 
#> [[3]]
#> [1] 1.732051

The latter allows registering different handlers for parallel execution, where we can use clustermq:

# set up the scheduler first, otherwise this will run sequentially

# this accepts same arguments as `Q`
clustermq::register_dopar_cmq(n_jobs=2, memory=1024)

# this will be executed as jobs
foreach(i=1:3) %dopar% sqrt(i)
#> Running sequentially ('LOCAL') ...
#> [[1]]
#> [1] 1
#> 
#> [[2]]
#> [1] 1.414214
#> 
#> [[3]]
#> [1] 1.732051

As BiocParallel supports foreach too, this means we can run all packages that use BiocParallel on the cluster as well via DoparParam.

library(BiocParallel)

# the number of jobs is ignored here since we're using the LOCAL scheduler
clustermq::register_dopar_cmq(n_jobs=2, memory=1024)
register(DoparParam())
bplapply(1:3, sqrt)

With targets

The targets package enables users to define a dependency structure of different function calls, and only evaluate them if the underlying data changed.

The targets package is a Make-like pipeline tool for statistics and data science in R. The package skips costly runtime for tasks that are already up to date, orchestrates the necessary computation with implicit parallel computing, and abstracts files as R objects. If all the current output matches the current upstream code and data, then the whole pipeline is up to date, and the results are more trustworthy than otherwise.

It can use clustermq to perform calculations as jobs.

LSF

Set the following options in your R session that will submit jobs:

options(
    clustermq.scheduler = "lsf",
    clustermq.template = "/path/to/file/below" # if using your own template
)

To supply your own template, save the contents below with any desired changes to a file and have clustermq.template point to it.

#BSUB-J {{ job_name }}[1-{{ n_jobs }}]  # name of the job / array jobs
#BSUB-n {{ cores | 1 }}                 # number of cores to use per job
#BSUB-o {{ log_file | /dev/null }}      # stdout + stderr; %I for array index
#BSUB-M {{ memory | 4096 }}             # Memory requirements in Mbytes
#BSUB-R rusage[mem={{ memory | 4096 }}] # Memory requirements in Mbytes
##BSUB-q default                        # name of the queue (uncomment)
##BSUB-W {{ walltime | 6:00 }}          # walltime (uncomment)

ulimit -v $(( 1024 * {{ memory | 4096 }} ))
CMQ_AUTH={{ auth }} R --no-save --no-restore -e 'clustermq:::worker("{{ master }}")'

In this file, #BSUB-* defines command-line arguments to the bsub program.

• Memory: defined by BSUB-M and BSUB-R. Check your local setup if the memory values supplied are MiB or KiB, default is 4096 if not requesting memory when calling Q()
• Queue: BSUB-q default. Use the queue with name default. This will most likely not exist on your system, so choose the right name and uncomment by removing the additional #
• Walltime: BSUB-W {{ walltime }}. Set the maximum time a job is allowed to run before being killed. The default here is to disable this line. If you enable it, enter a fixed value or pass the walltime argument to each function call. The way it is written, it will use 6 hours if no arguemnt is given.
• For other options, see the LSF documentation and add them via #BSUB-* (where * represents the argument)
• Do not change the identifiers in curly braces ({{ ... }}), as they are used to fill in the right variables

Once this is done, the package will use your settings and no longer warn you of the missing options.

SGE

Set the following options in your R session that will submit jobs:

options(
    clustermq.scheduler = "sge",
    clustermq.template = "/path/to/file/below" # if using your own template
)

To supply your own template, save the contents below with any desired changes to a file and have clustermq.template point to it.

#$ -N {{ job_name }}               # job name
##$ -q default                      # submit to queue named "default"
#$ -j y                            # combine stdout/error in one file
#$ -o {{ log_file | /dev/null }}   # output file
#$ -cwd                            # use pwd as work dir
#$ -V                              # use environment variable
#$ -t 1-{{ n_jobs }}               # submit jobs as array
#$ -pe smp {{ cores | 1 }}         # number of cores to use per job
#$ -l m_mem_free={{ memory | 1073741824 }} # 1 Gb in bytes

ulimit -v $(( 1024 * {{ memory | 4096 }} ))
CMQ_AUTH={{ auth }} R --no-save --no-restore -e 'clustermq:::worker("{{ master }}")'

In this file, #$-* defines command-line arguments to the qsub program.

• Queue: $ -q default. Use the queue with name default. This will most likely not exist on your system, so choose the right name and uncomment by removing the additional #
• For other options, see the SGE documentation. Do not change the identifiers in curly braces ({{ ... }}), as they are used to fill in the right variables.

Once this is done, the package will use your settings and no longer warn you of the missing options.

SLURM

Set the following options in your R session that will submit jobs:

options(
    clustermq.scheduler = "slurm",
    clustermq.template = "/path/to/file/below" # if using your own template
)

To supply your own template, save the contents below with any desired changes to a file and have clustermq.template point to it.

#!/bin/sh
#SBATCH --job-name={{ job_name }}
##SBATCH --partition=default
#SBATCH --output={{ log_file | /dev/null }}
#SBATCH --error={{ log_file | /dev/null }}
#SBATCH --mem-per-cpu={{ memory | 4096 }}
#SBATCH --array=1-{{ n_jobs }}
#SBATCH --cpus-per-task={{ cores | 1 }}

ulimit -v $(( 1024 * {{ memory | 4096 }} ))
CMQ_AUTH={{ auth }} R --no-save --no-restore -e 'clustermq:::worker("{{ master }}")'

In this file, #SBATCH defines command-line arguments to the sbatch program.

• Partition: SBATCH --partition default. Use the queue with name default. This will most likely not exist on your system, so choose the right name and uncomment by removing the additional #
• For other options, see the SLURM documentation. Do not change the identifiers in curly braces ({{ ... }}), as they are used to fill in the right variables.

Once this is done, the package will use your settings and no longer warn you of the missing options.

PBS

Set the following options in your R session that will submit jobs:

options(
    clustermq.scheduler = "pbs",
    clustermq.template = "/path/to/file/below" # if using your own template
)

To supply your own template, save the contents below with any desired changes to a file and have clustermq.template point to it.

#PBS -N {{ job_name }}
#PBS -J 1-{{ n_jobs }}
#PBS -l select=1:ncpus={{ cores | 1 }}:mpiprocs={{ cores | 1 }}:mem={{ memory | 4096 }}MB
#PBS -l walltime={{ walltime | 12:00:00 }}
#PBS -o {{ log_file | /dev/null }}
#PBS -j oe
##PBS -q default

ulimit -v $(( 1024 * {{ memory | 4096 }} ))
CMQ_AUTH={{ auth }} R --no-save --no-restore -e 'clustermq:::worker("{{ master }}")'

In this file, #PBS-* defines command-line arguments to the qsub program.

• Queue: #PBS-q default. Use the queue with name default. This will most likely not exist on your system, so choose the right name and uncomment by removing the additional #
• For other options, see the PBS documentation. Do not change the identifiers in curly braces ({{ ... }}), as they are used to fill in the right variables.

Once this is done, the package will use your settings and no longer warn you of the missing options.

Torque

Set the following options in your R session that will submit jobs:

options(
    clustermq.scheduler = "Torque",
    clustermq.template = "/path/to/file/below" # if using your own template
)

To supply your own template, save the contents below with any desired changes to a file and have clustermq.template point to it.

#PBS -N {{ job_name }}
#PBS -l nodes={{ n_jobs }}:ppn={{ cores | 1 }},walltime={{ walltime | 12:00:00 }}
#PBS -o {{ log_file | /dev/null }}
#PBS -j oe
##PBS -q default

ulimit -v $(( 1024 * {{ memory | 4096 }} ))
CMQ_AUTH={{ auth }} R --no-save --no-restore -e 'clustermq:::worker("{{ master }}")'

In this file, #PBS-* defines command-line arguments to the qsub program.

• Queue: #PBS-q default. Use the queue with name default. This will most likely not exist on your system, so choose the right name and uncomment by removing the additional #
• For other options, see the Torque documentation. Do not change the identifiers in curly braces ({{ ... }}), as they are used to fill in the right variables.

Once this is done, the package will use your settings and no longer warn you of the missing options.

SSH

While SSH is not a scheduler, we can access remote schedulers via SSH. If you want to use it, first make sure that clustermq works on your server with the real scheduler. Only then move on to setting up SSH.

options(
    clustermq.scheduler = "ssh",
    clustermq.ssh.host = "myhost", # set this up in your local ~/.ssh/config
    clustermq.ssh.log = "~/ssh_proxy.log",     # log file on your HPC
    clustermq.ssh.timeout = 30,    # if changing default connection timeout
    clustermq.template = "/path/to/file/below" # if using your own template
)

The default template is shown below. If R is not in your HPC $PATH, you may need to specify its path or load the required bash modules/conda environments.

To supply your own template, save its contents with any desired changes to a file on your local machine and have clustermq.template point to it.

ssh -o "ExitOnForwardFailure yes" -f
    -R {{ ctl_port }}:localhost:{{ local_port }}
    -R {{ job_port }}:localhost:{{ fwd_port }}
    {{ ssh_host }}
    "R --no-save --no-restore -e
        'clustermq:::ssh_proxy(ctl={{ ctl_port }}, job={{ job_port }})'
        > {{ ssh_log | /dev/null }} 2>&1"