User Guide

Installation

Install the clustermq package in R from CRAN (including the bundled ZeroMQ system library):

install.packages('clustermq')

Alternatively you can use the remotes package to install directly from Github:

# install.packages('remotes')
remotes::install_github('mschubert/clustermq')

In the develop branch, we will introduce code changes and new features. These may contain bugs, poor documentation, or other inconveniences. This branch may not install at times. However, feedback is very welcome.

# install.packages('remotes')
remotes::install_github('mschubert/clustermq', ref="develop")

Configuration

Choose your preferred parallelism using:

options(clustermq.scheduler = "your scheduler here")

There are three kinds of schedulers:

LOCAL - sequential processing of calls (default if no HPC scheduler found)
Multiple cores on the same machine
HPC schedulers

Parallel and HPC schedulers can also be used via SSH.

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 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.

Setting up the scheduler

An HPC cluster’s scheduler ensures that computing jobs are distributed to available worker nodes. Hence, this is what clustermq interfaces with in order to do computations.

By default, we will take whichever scheduler we find and fall back on local processing. This will work in most, but not all cases.

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

LSF - should work without setup
SGE - should work without setup
SLURM - should work without setup
PBS/Torque - needs options(clustermq.scheduler="PBS"/"Torque")
if you want another scheduler, open an issue

Default submission templates are provided and can be customized, e.g. to activate compute environments or containers.

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 in your ~/.Rprofile:

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.

Usage

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 pkg argument or referencing it with package_name:::

fx = function(x) {
    x %>%
        mutate(area = Sepal.Length * Sepal.Width) %>%
        head()
}
Q(fx, x=list(iris), pkgs="dplyr", n_jobs=1)
#> Running sequentially ('LOCAL') ...
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
#> [[1]]
#>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species  area
#> 1          5.1         3.5          1.4         0.2  setosa 17.85
#> 2          4.9         3.0          1.4         0.2  setosa 14.70
#> 3          4.7         3.2          1.3         0.2  setosa 15.04
#> 4          4.6         3.1          1.5         0.2  setosa 14.26
#> 5          5.0         3.6          1.4         0.2  setosa 18.00
#> 6          5.4         3.9          1.7         0.4  setosa 21.06

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)
x = foreach(i=1:3) %do% sqrt(i)

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

x = foreach(i=1:3) %dopar% sqrt(i)
#> Warning: executing %dopar% sequentially: no parallel backend registered

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

# set up the scheduler first, otherwise this will run sequentially
clustermq::register_dopar_cmq(n_jobs=2, memory=1024) # this accepts same arguments as `Q`
x = foreach(i=1:3) %dopar% sqrt(i) # this will be executed as jobs
#> Running sequentially ('LOCAL') ...

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

library(BiocParallel)
register(DoparParam()) # after register_dopar_cmq(...)
bplapply(1:3, sqrt)

With `drake`

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

drake — or, Data Frames in R for Make — is a general-purpose workflow manager for data-driven tasks. It rebuilds intermediate data objects when their dependencies change, and it skips work when the results are already up to date. Not every runthrough starts from scratch, and completed workflows have tangible evidence of reproducibility. drake also supports scalability, parallel computing, and a smooth user experience when it comes to setting up, deploying, and maintaining data science projects.

It can use clustermq to perform calculations as jobs:

library(drake)
load_mtcars_example()
# clean(destroy = TRUE)
# options(clustermq.scheduler = "multicore")
make(my_plan, parallelism = "clustermq", jobs = 2, verbose = 4)

Options

The various configurable options are mentioned throughout the documentation, where applicable, however, we list all of the options here for reference.

Options can be set by including a call to options(<key> = <value>) in your .Rprofile, or by calling options(<key> = <value>) in a script or interactively during a session.

clustermq.scheduler - One of the supported clustermq schedulers; options are "LOCAL", "multiprocess", "multicore", "lsf", "sge", "slurm", "pbs", "Torque", or "ssh" (default is the HPC scheduler found in $PATH, otherwise "LOCAL")
clustermq.host - The name of the node or device for constructing the ZeroMQ host address (default is Sys.info()["nodename"])
clustermq.ssh.host - The user name and host for connecting to the HPC via SSH (e.g. user@host); we recommend setting up SSH keys for password-less login
clustermq.ssh.log - Path for a file (on the SSH host) that will be created and populated with logging information regarding the SSH connection (e.g. "~/cmq_ssh.log"); helpful for debugging purposes
clustermq.ssh.timeout - The amount of time to wait (in seconds) for a SSH start-up connection before timing out (default is 5 seconds)
clustermq.worker.timeout - The amount of time to wait (in seconds) for master-worker communication before timing out (default is 600 seconds)
clustermq.error.timeout - The amount of time to wait (in seconds), in case of a worker error, for the remaining workers to finish their computations and shut down cleanly (default is min(timeout, 30) seconds)
clustermq.template - Path to a template file for submitting HPC jobs; only necessary if using your own template, otherwise the default template will be used (default depends on clustermq.scheduler)
clustermq.data.warning - The threshold for the size of the common data (in Mb) before clustermq throws a warning (default is 1000)
clustermq.defaults - A named-list of default values for the HPC template; this takes precedence over defaults specified in the template file (default is an empty list (i.e. list()))

Troubleshooting

Debugging workers

Function calls evaluated by workers are wrapped in event handlers, which means that even if a call evaluation throws an error, this should be reported back to the main R session.

However, there are reasons why workers might crash, and in which case they can not report back. These include:

A segfault in a low-level process
Process kill due to resource constraints (e.g. walltime)
Reaching the wait timeout without any signal from the master process
Probably others

In this case, it is useful to have the worker(s) create a log file that will also include events that are not reported back. It can be requested using:

Q(..., log_worker=TRUE)

This will create a file called -.log in your current working directory, irrespective of which scheduler you use.

You can customize the file name using

Q(..., template=list(log_file = <yourlog>))

Note that in this case log_file is a template field of your scheduler script, and hence needs to be present there in order for this to work. The default templates all have this field included.

In order to log each worker separately, some schedulers support wildcards in their log file names. For instance:

Multicore/Multiprocess: log_file="/path/to.file.%i"
SGE: log_file="/path/to.file.\$TASK_ID"
LSF: log_file="/path/to.file.%I"
Slurm: log_file="/path/to.file.%a"
PBS: log_file="/path/to.file.$PBS_ARRAY_INDEX"
Torque: log_file="/path/to.file.$PBS_ARRAYID"

Your scheduler documentation will have more details about the available options.

When reporting a bug that includes worker crashes, please always include a log file.

SSH

Before trying remote schedulers via SSH, make sure that the scheduler works when you first connect to the cluster and run a job from there.

If the terminal is stuck at

Connecting <user@host> via SSH ...

make sure that each step of your SSH connection works by typing the following commands in your local terminal and make sure that you don’t get errors or warnings in each step:

# test your ssh login that you set up in ~/.ssh/config
# if this fails you have not set up SSH correctly
ssh <user@host>

# test port forwarding from 54709 remote to 6687 local (ports are random)
# if the fails you will not be able to use clustermq via SSH
ssh -R 54709:localhost:6687 <user@host> R --vanilla

If you get an Command not found: R error, make sure your $PATH is set up correctly in your ~/.bash_profile and/or your ~/.bashrc (depending on your cluster config you might need either). You may also need to modify your SSH template to load R as a module or conda environment.

If you get a SSH warning or error try again with ssh -v to enable verbose output. If the forward itself works, run the following in your local R session (ideally also in command-line R, not only in RStudio):

options(clustermq.scheduler = "ssh",
        clustermq.ssh.log = "~/ssh_proxy.log")
Q(identity, x=1, n_jobs=1)

This will create a log file on the remote server that will contain any errors that might have occurred during ssh_proxy startup.

If the ssh_proxy startup fails on your local machine with the error

Remote R process did not respond after 5 seconds. Check your SSH server log.

but the server log does not show any errors, then you can try increasing the timeout:

options(clustermq.ssh.timeout = 30) # in seconds

This can happen when your SSH startup template includes additional steps before starting R, such as activating a module or conda environment, or having to confirm the connection via two-factor authentication.

Environments

Environments for workers

In some cases, it may be necessary to activate a specific computing environment on the scheduler jobs prior to starting up the worker. This can be, for instance, because R was only installed in a specific environment or container.

Examples for such environments or containers are:

Bash module environments
Conda environments
Docker/Singularity containers

It should be possible to activate them in the job submission script (i.e., the template file). This is widely untested, but would look the following for the LSF scheduler (analogous for others):

#BSUB-J {{ job_name }}[1-{{ n_jobs }}]  # name of the job / array jobs
#BSUB-o {{ log_file | /dev/null }}      # stdout + stderr
#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)

module load {{ bashenv | default_bash_env }}
# or: source activate {{ conda | default_conda_env_name }}
# or: your environment activation command
ulimit -v $(( 1024 * {{ memory | 4096 }} ))
CMQ_AUTH={{ auth }} R --no-save --no-restore -e 'clustermq:::worker("{{ master }}")'

This template still needs to be filled, so in the above example you need to pass either

Q(..., template=list(bashenv="my environment name"))

or set it via an .Rprofile option:

options(
    clustermq.defaults = list(bashenv="my default env")
)

Running master inside containers

If your master process is inside a container, accessing the HPC scheduler is more difficult. Containers, including singularity and docker, isolate the processes inside the container from the host. The R process will not be able to submit a job because the scheduler cannot be found.

Note that the HPC node running the master process must be allowed to submit jobs. Not all HPC systems allow compute nodes to submit jobs. If that is the case, you may need to run the master process on the login node, and discuss the issue with your system administrator.

If your container is binary compatible with the host, you may be able to bind in the scheduler executable to the container.

For example, PBS might look something like:

#PBS directives ...

module load singularity

SINGULARITYENV_APPEND_PATH=/opt/pbs/bin
singularity exec --bind /opt/pbs/bin r_image.sif Rscript master_script.R

A working example of binding SLURM into a CentOS 7 container image from a CentOS 7 host is available at https://groups.google.com/a/lbl.gov/d/msg/singularity/syLcsIWWzdo/NZvF2Ud2AAAJ

Alternatively, you can create a script that uses SSH to execute the scheduler on the login node. For this, you will need an SSH client in the container, keys set up for password-less login, and create a script to call the scheduler on the login node via ssh (e.g. ~/bin/qsub for SGE/PBS/Torque, bsub for LSF and sbatch for Slurm):

#!/bin/bash
ssh -i ~/.ssh/<your key file> ${PBS_O_HOST:-"no_host_not_in_a_pbs_job"} qsub "$@"

Make sure the script is executable, and bind/copy it into the container somewhere on $PATH. Home directories are bound in by default in singularity.

chmod u+x ~/bin/qsub
SINGULARITYENV_APPEND_PATH=~/bin

Scheduler templates

LSF

In your ~/.Rprofile on your computing cluster, set the following options:

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

The option clustermq.template should point to a LSF template file like the one below (only needed if you want to supply your own template rather than using the default).

#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 (or comment out this line with an 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

In your ~/.Rprofile on your computing cluster, set the following options:

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

The option clustermq.template should point to a SGE template file like the one below (only needed if you want to supply your own template rather than using the default).

#$ -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 {{ cores | 1 }}             # number of cores to use per job

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 (or comment out this line with an 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

In your ~/.Rprofile on your computing cluster, set the following options:

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

The option clustermq.template should point to a SLURM template file like the one below (only needed if you want to supply your own template rather than using the default).

#!/bin/sh
#SBATCH --job-name={{ job_name }}
#SBATCH --partition=default
#SBATCH --output={{ log_file | /dev/null }} # you can add .%a for array index
#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.

Queue: SBATCH --partition default. Use the queue with name default. This will most likely not exist on your system, so choose the right name (or comment out this line with an 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

In your ~/.Rprofile on your computing cluster, set the following options:

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

The option clustermq.template should point to a PBS template file like the one below (only needed if you want to supply your own template rather than using the default).

#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 (or comment out this line with an 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

In your ~/.Rprofile on your computing cluster, set the following options:

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

The option clustermq.template should point to a Torque template file like the one below (only needed if you want to supply your own template rather than using the default).

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

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 (or comment out this line with an 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 your real scheduler is running when manually connecting to the HPC environment.

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 the 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.

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"

Installation

Configuration

Local parallelization

Setting up the scheduler

SSH connector

Usage

The Q function

Examples

As parallel foreach backend

With drake

Options

Troubleshooting

Debugging workers

SSH

Environments

Environments for workers

Running master inside containers

Scheduler templates

LSF

SGE

SLURM

PBS

Torque

SSH

The `Q` function

As parallel `foreach` backend

With `drake`