Package 'parallelly'

Description

Coerce an Object to a Cluster Object

Usage

as.cluster(x, ...)

## S3 method for class 'cluster'
as.cluster(x, ...)

## S3 method for class 'list'
as.cluster(x, ...)

## S3 method for class 'SOCKnode'
as.cluster(x, ...)

## S3 method for class 'SOCK0node'
as.cluster(x, ...)

## S3 method for class 'RichSOCKnode'
as.cluster(x, ...)

## S3 method for class 'cluster'
c(..., recursive = FALSE)

Arguments

Value

An object of class cluster.

c(...) combine multiple clusters and / or cluster nodes into one cluster returned as an of class cluster. A warning will be produced if there are duplicated nodes in the resulting cluster.

Examples

cl1 <- makeClusterPSOCK(2, dryrun = TRUE)
cl2 <- makeClusterPSOCK(c("n1", "server.remote.org"), dryrun = TRUE)
cl <- c(cl1, cl2)
print(cl)

Description

Registers a finalizer to a cluster such that the cluster will be stopped when garbage collected

Usage

autoStopCluster(cl, debug = FALSE)

Arguments

Details

The cluster is stopped using stopCluster(cl). An alternative to explicitly call this function on an existing cluster object, is to create the cluster object using makeClusterPSOCK() with argument autoStop = TRUE.

Value

The cluster object with attribute gcMe set.

Examples

cl <- makeClusterPSOCK(2, dryrun = TRUE)
cl <- autoStopCluster(cl)
print(cl)
rm(list = "cl")
gc()

Description

The number of connections that can be open at the same time in R is typically 128, where the first three are occupied by the always open stdin(), stdout(), and stderr() connections, which leaves 125 slots available for other types of connections. Connections are used in many places, e.g. reading and writing to file, downloading URLs, communicating with parallel R processes over a socket connections (e.g. parallel::makeCluster() and makeClusterPSOCK()), and capturing standard output via text connections (e.g. utils::capture.output()).

Usage

availableConnections()

freeConnections()

Value

A non-negative integer, or +Inf if the available number of connections is greater than 16384, which is a limit be set via option parallelly.availableConnections.tries.

How to increase the limit

In R (>= 4.4.0), it is possible to increase the limit of 128 connections to a greater number via command-line option --max-connections=N, e.g.

$ R --max-connection=512

For R (< 4.4.0), the limit can only be changed by rebuilding R from source, because the limited is hardcoded as a

#define NCONNECTIONS 128

in ‘src/main/connections.c’.

How the limit is identified

Since the limit might changed, for instance in custom R builds or in future releases of R, we do not want to assume that the limit is 128 for all R installation. Unfortunately, it is not possible to query R for what the limit is. Instead, availableConnections() infers it from trial-and-error. until it fails. For efficiency, the result is memoized throughout the current R session.

References

1. 'WISH: Increase limit of maximum number of open connections (currently 125+3)', 2016-07-09, https://github.com/HenrikBengtsson/Wishlist-for-R/issues/28

See Also

base::showConnections().

Examples

total <- availableConnections()
message("You can have ", total, " connections open in this R installation")
free <- freeConnections()
message("There are ", free, " connections remaining")

Description

The current/main R session counts as one, meaning the minimum number of cores available is always at least one.

Usage

availableCores(
  constraints = NULL,
  methods = getOption2("parallelly.availableCores.methods", c("system", "cgroups.cpuset",
    "cgroups.cpuquota", "cgroups2.cpu.max", "nproc", "mc.cores", "BiocParallel",
    "_R_CHECK_LIMIT_CORES_", "Bioconductor", "LSF", "PJM", "PBS", "SGE", "Slurm",
    "fallback", "custom")),
  na.rm = TRUE,
  logical = getOption2("parallelly.availableCores.logical", TRUE),
  default = c(current = 1L),
  which = c("min", "max", "all"),
  omit = getOption2("parallelly.availableCores.omit", 0L)
)

Arguments

Details

The following settings ("methods") for inferring the number of cores are supported:

• "system" - Query detectCores(logical = logical).

• "cgroups.cpuset" - On Unix, query control group (cgroup) value cpuset.set.

• "cgroups.cpuquota" - On Unix, query control group (cgroup) value cpu.cfs_quota_us / cpu.cfs_period_us.

• "cgroups2.cpu.max" - On Unix, query control group (cgroup v2) values cpu.max.

• "nproc" - On Unix, query system command nproc.

• "mc.cores" - If available, returns the value of option mc.cores. Note that mc.cores is defined as the number of additional R processes that can be used in addition to the main R process. This means that with mc.cores = 0 all calculations should be done in the main R process, i.e. we have exactly one core available for our calculations. The mc.cores option defaults to environment variable MC_CORES (and is set accordingly when the parallel package is loaded). The mc.cores option is used by for instance mclapply() of the parallel package.

• "connections" - Query the current number of available R connections per freeConnections(). This is the maximum number of socket-based parallel cluster nodes that are possible launch, because each one needs its own R connection. The exception is when freeConnections() is zero, then 1L is still returned, because availableCores() should always return a positive integer.

• "BiocParallel" - Query environment variable BIOCPARALLEL_WORKER_NUMBER (integer), which is defined and used by BiocParallel (>= 1.27.2). If the former is set, this is the number of cores considered.

• "_R_CHECK_LIMIT_CORES_" - Query environment variable _R_CHECK_LIMIT_CORES_ (logical or "warn") used by ⁠R CMD check⁠ and set to true by ⁠R CMD check --as-cran⁠. If set to a non-false value, then a maximum of 2 cores is considered.

• "Bioconductor" - Query environment variable IS_BIOC_BUILD_MACHINE (logical) used by the Bioconductor (>= 3.16) build and check system. If set to true, then a maximum of 4 cores is considered.

• "LSF" - Query Platform Load Sharing Facility (LSF) environment variable LSB_DJOB_NUMPROC. Jobs with multiple (CPU) slots can be submitted on LSF using ⁠bsub -n 2 -R "span[hosts=1]" < hello.sh⁠.

• "PJM" - Query Fujitsu Technical Computing Suite (that we choose to shorten as "PJM") environment variables PJM_VNODE_CORE and PJM_PROC_BY_NODE. The first is set when submitted with ⁠pjsub -L vnode-core=8 hello.sh⁠.

• "PBS" - Query TORQUE/PBS environment variables PBS_NUM_PPN and NCPUS. Depending on PBS system configuration, these resource parameters may or may not default to one. An example of a job submission that results in this is ⁠qsub -l nodes=1:ppn=2⁠, which requests one node with two cores.

• "SGE" - Query Sun Grid Engine/Oracle Grid Engine/Son of Grid Engine (SGE) and Univa Grid Engine (UGE) environment variable NSLOTS. An example of a job submission that results in this is ⁠qsub -pe smp 2⁠ (or ⁠qsub -pe by_node 2⁠), which requests two cores on a single machine.

• "Slurm" - Query Simple Linux Utility for Resource Management (Slurm) environment variable SLURM_CPUS_PER_TASK. This may or may not be set. It can be set when submitting a job, e.g. ⁠sbatch --cpus-per-task=2 hello.sh⁠ or by adding ⁠#SBATCH --cpus-per-task=2⁠ to the ‘hello.sh’ script. If SLURM_CPUS_PER_TASK is not set, then it will fall back to use SLURM_CPUS_ON_NODE if the job is a single-node job (SLURM_JOB_NUM_NODES is 1), e.g. ⁠sbatch --ntasks=2 hello.sh⁠. To make sure all tasks are assign to a single node, specify --nodes=1, e.g. ⁠sbatch --nodes=1 --ntasks=16 hello.sh⁠.

• "custom" - If option parallelly.availableCores.custom is set and a function, then this function will be called (without arguments) and it's value will be coerced to an integer, which will be interpreted as a number of available cores. If the value is NA, then it will be ignored. It is safe for this custom function to call availableCores(); if done, the custom function will not be recursively called.

For any other value of a methods element, the R option with the same name is queried. If that is not set, the system environment variable is queried. If neither is set, a missing value is returned.

Value

Return a positive (>= 1) integer. If which = "all", then more than one value may be returned. Together with na.rm = FALSE missing values may also be returned.

Avoid ending up with zero cores

Note that some machines might have a limited number of cores, or the R process runs in a container or a cgroup that only provides a small number of cores. In such cases:

ncores <- availableCores() - 1

may return zero, which is often not intended and is likely to give an error downstream. Instead, use:

ncores <- availableCores(omit = 1)

to put aside one of the cores from being used. Regardless how many cores you put aside, this function is guaranteed to return at least one core.

Advanced usage

It is possible to override the maximum number of cores on the machine as reported by availableCores(methods = "system"). This can be done by first specifying options(parallelly.availableCores.methods = "mc.cores") and then the number of cores to use, e.g. options(mc.cores = 8).

See Also

To get the set of available workers regardless of machine, see availableWorkers().

Examples

message(paste("Number of cores available:", availableCores()))

## Not run: 
options(mc.cores = 2L)
message(paste("Number of cores available:", availableCores()))

## End(Not run)

## Not run: 
## IMPORTANT: availableCores() may return 1L
options(mc.cores = 1L)
ncores <- availableCores() - 1      ## ncores = 0
ncores <- availableCores(omit = 1)  ## ncores = 1
message(paste("Number of cores to use:", ncores))

## End(Not run)

## Not run: 
## Use 75% of the cores on the system but never more than four
options(parallelly.availableCores.custom = function() {
  ncores <- max(parallel::detectCores(), 1L, na.rm = TRUE)
  ncores <- min(as.integer(0.75 * ncores), 4L)
  max(1L, ncores)
})
message(paste("Number of cores available:", availableCores()))

## Use 50% of the cores according to availableCores(), e.g.
## allocated by a job scheduler or cgroups.
## Note that it is safe to call availableCores() here.
options(parallelly.availableCores.custom = function() {
  0.50 * parallelly::availableCores()
})
message(paste("Number of cores available:", availableCores()))

## End(Not run)

Description

Get Set of Available Workers

Usage

availableWorkers(
  constraints = NULL,
  methods = getOption2("parallelly.availableWorkers.methods", c("mc.cores",
    "BiocParallel", "_R_CHECK_LIMIT_CORES_", "Bioconductor", "LSF", "PJM", "PBS", "SGE",
    "Slurm", "custom", "cgroups.cpuset", "cgroups.cpuquota", "cgroups2.cpu.max", "nproc",
    "system", "fallback")),
  na.rm = TRUE,
  logical = getOption2("parallelly.availableCores.logical", TRUE),
  default = getOption2("parallelly.localhost.hostname", "localhost"),
  which = c("auto", "min", "max", "all")
)

Arguments

Details

The default set of workers for each method is rep("localhost", times = availableCores(methods = method, logical = logical)), which means that each will at least use as many parallel workers on the current machine that availableCores() allows for that method.

In addition, the following settings ("methods") are also acknowledged:

• "LSF" - Query LSF/OpenLava environment variable LSB_HOSTS.

• "PJM" - Query Fujitsu Technical Computing Suite (that we choose to shorten as "PJM") the hostname file given by environment variable PJM_O_NODEINF. The PJM_O_NODEINF file lists the hostnames of the nodes allotted. This function returns those hostnames each repeated availableCores() times, where availableCores() reflects PJM_VNODE_CORE. For example, for ⁠pjsub -L vnode=2 -L vnode-core=8 hello.sh⁠, the PJM_O_NODEINF file gives two hostnames, and PJM_VNODE_CORE gives eight cores per host, resulting in a character vector of 16 hostnames (for two unique hostnames).

• "PBS" - Query TORQUE/PBS environment variable PBS_NODEFILE. If this is set and specifies an existing file, then the set of workers is read from that file, where one worker (node) is given per line. An example of a job submission that results in this is ⁠qsub -l nodes=4:ppn=2⁠, which requests four nodes each with two cores.

• "SGE" - Query Sun Grid Engine/Oracle Grid Engine/Son of Grid Engine (SGE) and Univa Grid Engine (UGE) environment variable PE_HOSTFILE. An example of a job submission that results in this is ⁠qsub -pe mpi 8⁠ (or ⁠qsub -pe ompi 8⁠), which requests eight cores on a any number of machines.

• "Slurm" - Query Slurm environment variable SLURM_JOB_NODELIST (fallback to legacy SLURM_NODELIST) and parse set of nodes. Then query Slurm environment variable SLURM_JOB_CPUS_PER_NODE (fallback SLURM_TASKS_PER_NODE) to infer how many CPU cores Slurm have allotted to each of the nodes. If SLURM_CPUS_PER_TASK is set, which is always a scalar, then that is respected too, i.e. if it is smaller, then that is used for all nodes. For example, if SLURM_NODELIST="n1,n[03-05]" (expands to c("n1", "n03", "n04", "n05")) and SLURM_JOB_CPUS_PER_NODE="2(x2),3,2" (expands to c(2, 2, 3, 2)), then c("n1", "n1", "n03", "n03", "n04", "n04", "n04", "n05", "n05") is returned. If in addition, SLURM_CPUS_PER_TASK=1, which can happen depending on hyperthreading configurations on the Slurm cluster, then c("n1", "n03", "n04", "n05") is returned.

• "custom" - If option parallelly.availableWorkers.custom is set and a function, then this function will be called (without arguments) and it's value will be coerced to a character vector, which will be interpreted as hostnames of available workers. It is safe for this custom function to call availableWorkers(); if done, the custom function will not be recursively called.

Value

Return a character vector of workers, which typically consists of names of machines / compute nodes, but may also be IP numbers.

Known limitations

availableWorkers(methods = "Slurm") will expand SLURM_JOB_NODELIST using scontrol show hostnames "$SLURM_JOB_NODELIST", if available. If not available, then it attempts to parse the compressed nodelist based on a best-guess understanding on what the possible syntax may be. One known limitation is that "multi-dimensional" ranges are not supported, e.g. "a[1-2]b[3-4]" is expanded by scontrol to c("a1b3", "a1b4", "a2b3", "a2b4"). If scontrol is not available, then any components that failed to be parsed are dropped with an informative warning message. If no components could be parsed, then the result of methods = "Slurm" will be empty.

See Also

To get the number of available workers on the current machine, see availableCores().

Examples

message(paste("Available workers:",
        paste(sQuote(availableWorkers()), collapse = ", ")))

## Not run: 
options(mc.cores = 2L)
message(paste("Available workers:",
        paste(sQuote(availableWorkers()), collapse = ", ")))

## End(Not run)

## Not run: 
## Always use two workers on host 'n1' and one on host 'n2'
options(parallelly.availableWorkers.custom = function() {
  c("n1", "n1", "n2")
})
message(paste("Available workers:",
        paste(sQuote(availableWorkers()), collapse = ", ")))

## End(Not run)

## Not run: 
## A 50% random subset of the available workers.
## Note that it is safe to call availableWorkers() here.
options(parallelly.availableWorkers.custom = function() {
  workers <- parallelly::availableWorkers()
  sample(workers, size = 0.50 * length(workers))
})
message(paste("Available workers:",
        paste(sQuote(availableWorkers()), collapse = ", ")))

## End(Not run)

Description

The makeClusterPSOCK() function creates a cluster of R workers for parallel processing. These R workers may be background R sessions on the current machine, R sessions on external machines (local or remote), or a mix of such. For external workers, the default is to use SSH to connect to those external machines. This function works similarly to makePSOCKcluster() of the parallel package, but provides additional and more flexibility options for controlling the setup of the system calls that launch the background R workers, and how to connect to external machines.

Usage

checkNumberOfLocalWorkers(workers)

makeClusterPSOCK(
  workers,
  makeNode = makeNodePSOCK,
  port = c("auto", "random"),
  ...,
  autoStop = FALSE,
  tries = getOption2("parallelly.makeNodePSOCK.tries", 3L),
  delay = getOption2("parallelly.makeNodePSOCK.tries.delay", 15),
  validate = getOption2("parallelly.makeNodePSOCK.validate", TRUE),
  verbose = getOption2("parallelly.debug", FALSE)
)

makeNodePSOCK(
  worker = getOption2("parallelly.localhost.hostname", "localhost"),
  master = NULL,
  port,
  connectTimeout = getOption2("parallelly.makeNodePSOCK.connectTimeout", 2 * 60),
  timeout = getOption2("parallelly.makeNodePSOCK.timeout", 30 * 24 * 60 * 60),
  rscript = NULL,
  homogeneous = NULL,
  rscript_args = NULL,
  rscript_envs = NULL,
  rscript_libs = NULL,
  rscript_startup = NULL,
  rscript_sh = c("auto", "cmd", "sh"),
  default_packages = c("datasets", "utils", "grDevices", "graphics", "stats", if
    (methods) "methods"),
  methods = TRUE,
  socketOptions = getOption2("parallelly.makeNodePSOCK.socketOptions", "no-delay"),
  useXDR = getOption2("parallelly.makeNodePSOCK.useXDR", FALSE),
  outfile = "/dev/null",
  renice = NA_integer_,
  rshcmd = getOption2("parallelly.makeNodePSOCK.rshcmd", NULL),
  user = NULL,
  revtunnel = NA,
  rshlogfile = NULL,
  rshopts = getOption2("parallelly.makeNodePSOCK.rshopts", NULL),
  rank = 1L,
  manual = FALSE,
  dryrun = FALSE,
  quiet = FALSE,
  setup_strategy = getOption2("parallelly.makeNodePSOCK.setup_strategy", "parallel"),
  action = c("launch", "options"),
  verbose = FALSE
)

Arguments

Value

An object of class c("RichSOCKcluster", "SOCKcluster", "cluster") consisting of a list of "SOCKnode" or "SOCK0node" workers (that also inherit from RichSOCKnode).

makeNodePSOCK() returns a "SOCKnode" or "SOCK0node" object representing an established connection to a worker.

Protection against CPU overuse

Using too many parallel workers on the same machine may result in overusing the CPU. For example, if an R script hard codes the number of parallel workers to 32, as in

cl <- makeClusterPSOCK(32)

it will use more than 100% of the CPU cores when running on machine with fewer than 32 CPU cores. For example, on a eight-core machine, this may run the CPU at 400% of its capacity, which has a significant negative effect on the current R process, but also on all other processes running on the same machine. This also a problem on systems where R gets allotted a specific number of CPU cores, which is the case on high-performance compute (HPC) clusters, but also on other shared systems that limits user processes via Linux Control Groups (CGroups). For example, a free account on Posit Cloud is limited to a single CPU core. Parallelizing with 32 workers when only having access to a single core, will result in 3200% overuse and 32 concurrent R processes competing for this single CPU core.

To protect against CPU overuse by mistake, makeClusterPSOCK() will warn when parallelizing above 100%;

cl <- parallelly:::makeClusterPSOCK(12, dryrun = TRUE)
Warning message:
In checkNumberOfLocalWorkers(workers) :
  Careful, you are setting up 12 localhost parallel workers with
only 8 CPU cores available for this R process, which could result
in a 150% load. The maximum is set to 100%. Overusing the CPUs has
negative impact on the current R process, but also on all other
processes of yours and others running on the same machine. See
help("parallelly.options", package = "parallelly") for how to
override this threshold

Any attempts resulting in more than 300% overuse will be refused;

> cl <- parallelly:::makeClusterPSOCK(25, dryrun = TRUE)
Error in checkNumberOfLocalWorkers(workers) : 
  Attempting to set up 25 localhost parallel workers with only
8 CPU cores available for this R process, which could result in
a 312% load. The maximum is set to 300%. Overusing the CPUs has
negative impact on the current R process, but also on all other
processes of yours and others running on the same machine. See
help("parallelly.options", package = "parallelly") for how to
override this threshold

See parallelly.options for how to change the default thresholds.

Definition of localhost

A hostname is considered to be localhost if it equals:

• "localhost",

• "127.0.0.1", or

• Sys.info()[["nodename"]].

It is also considered localhost if it appears on the same line as the value of Sys.info()[["nodename"]] in file ‘/etc/hosts’.

Default SSH client and options (arguments rshcmd and rshopts)

Arguments rshcmd and rshopts are only used when connecting to an external host.

The default method for connecting to an external host is via SSH and the system executable for this is given by argument rshcmd. The default is given by option parallelly.makeNodePSOCK.rshcmd. If that is not set, then the default is to use ssh on Unix-like systems, including macOS as well as Windows 10. On older MS Windows versions, which does not have a built-in ssh client, the default is to use (i) plink from the PuTTY project, and then (ii) the ssh client that is distributed with RStudio.

PuTTY puts itself on Windows' system PATH when installed, meaning this function will find PuTTY automatically if installed. If not, to manually set specify PuTTY as the SSH client, specify the absolute pathname of ‘plink.exe’ in the first element and option -ssh in the second as in rshcmd = c("C:/Path/PuTTY/plink.exe", "-ssh"). This is because all elements of rshcmd are individually "shell" quoted and element rshcmd[1] must be on the system PATH.

Furthermore, when running R from RStudio on Windows, the ssh client that is distributed with RStudio will also be considered. This client, which is from MinGW MSYS, is searched for in the folder given by the RSTUDIO_MSYS_SSH environment variable—a variable that is (only) set when running RStudio. To use this SSH client outside of RStudio, set RSTUDIO_MSYS_SSH accordingly.

You can override the default set of SSH clients that are searched for by specifying them in argument rshcmd or via option parallelly.makeNodePSOCK.rshcmd using the format ⁠<...>⁠, e.g. rshcmd = c("<rstudio-ssh>", "<putty-plink>", "<ssh>"). See below for examples.

If no SSH-client is found, an informative error message is produced.

Additional SSH command-line options may be specified via argument rshopts, which defaults to option parallelly.makeNodePSOCK.rshopts. For instance, a private SSH key can be provided as rshopts = c("-i", "~/.ssh/my_private_key"). PuTTY users should specify a PuTTY PPK file, e.g. rshopts = c("-i", "C:/Users/joe/.ssh/my_keys.ppk"). Contrary to rshcmd, elements of rshopts are not quoted.

Accessing external machines that prompts for a password

IMPORTANT: With one exception, it is not possible to for these functions to log in and launch R workers on external machines that requires a password to be entered manually for authentication. The only known exception is the PuTTY client on Windows for which one can pass the password via command-line option -pw, e.g. rshopts = c("-pw", "MySecretPassword").

Note, depending on whether you run R in a terminal or via a GUI, you might not even see the password prompt. It is also likely that you cannot enter a password, because the connection is set up via a background system call.

The poor man's workaround for setup that requires a password is to manually log into the each of the external machines and launch the R workers by hand. For this approach, use manual = TRUE and follow the instructions which include cut'n'pasteable commands on how to launch the worker from the external machine.

However, a much more convenient and less tedious method is to set up key-based SSH authentication between your local machine and the external machine(s), as explain below.

Accessing external machines via key-based SSH authentication

The best approach to automatically launch R workers on external machines over SSH is to set up key-based SSH authentication. This will allow you to log into the external machine without have to enter a password.

Key-based SSH authentication is taken care of by the SSH client and not R. To configure this, see the manuals of your SSH client or search the web for "ssh key authentication".

Reverse SSH tunneling

If SSH is used, which is inferred from rshcmd[1], then the default is to use reverse SSH tunneling (revtunnel = TRUE), otherwise not (revtunnel = FALSE). Using reverse SSH tunneling, avoids complications from otherwise having to configure port forwarding in firewalls, which often requires static IP address as well as privileges to edit the firewall on your outgoing router, something most users don't have. It also has the advantage of not having to know the internal and / or the public IP address / hostname of the master. Yet another advantage is that there will be no need for a DNS lookup by the worker machines to the master, which may not be configured or is disabled on some systems, e.g. compute clusters.

Argument rscript

If homogeneous is FALSE, the rscript defaults to "Rscript", i.e. it is assumed that the Rscript executable is available on the PATH of the worker. If homogeneous is TRUE, the rscript defaults to file.path(R.home("bin"), "Rscript"), i.e. it is basically assumed that the worker and the caller share the same file system and R installation.

When specified, argument rscript should be a character vector with one or more elements. Any asterisk ("*") will be resolved to the above default homogeneous-dependent Rscript path. All elements are automatically shell quoted using base::shQuote(), except those that are of format ⁠<ENVVAR>=<VALUE>⁠, that is, the ones matching the regular expression '‘⁠^[[:alpha:]_][[:alnum:]_]*=.*⁠’'. Another exception is when rscript inherits from 'AsIs'.

Default value of argument homogeneous

The default value of homogeneous is TRUE if and only if either of the following is fulfilled:

• worker is localhost

• revtunnel is FALSE and master is localhost

• worker is neither an IP number nor a fully qualified domain name (FQDN). A hostname is considered to be a FQDN if it contains one or more periods

In all other cases, homogeneous defaults to FALSE.

Connection timeout

Argument connectTimeout does not work properly on Unix and macOS due to limitation in R itself. For more details on this, please see R-devel thread 'BUG?: On Linux setTimeLimit() fails to propagate timeout error when it occurs (works on Windows)' on 2016-10-26 (https://stat.ethz.ch/pipermail/r-devel/2016-October/073309.html). When used, the timeout will eventually trigger an error, but it won't happen until the socket connection timeout timeout itself happens.

Communication timeout

If there is no communication between the master and a worker within the timeout limit, then the corresponding socket connection will be closed automatically. This will eventually result in an error in code trying to access the connection. This timeout is also what terminates a stray-running parallel cluster-node process.

Failing to set up local workers

When setting up a cluster of localhost workers, that is, workers running on the same machine as the master R process, occasionally a connection to a worker ("cluster node") may fail to be set up. When this occurs, an informative error message with troubleshooting suggestions will be produced. The most common reason for such localhost failures is due to port clashes. Retrying will often resolve the problem.

If R stalls when setting up a cluster of local workers, then it might be that you have a virtual private network (VPN) enabled that is configured to prevent you from connecting to localhost. To verify that this is the case, call the following from the terminal:

{local}$ ssh localhost "date"

This also freezed if the VPN intercepts connections to localhost. If this happens, try also:

{local}$ ssh 127.0.0.1 "date"

In rare cases, ⁠127.0.0.1⁠ might work when localhost does not. If the latter works, setting R option:

options(parallelly.localhost.hostname = "127.0.0.1")

should solve it (the default is "localhost"). You can set this automatically when R starts by adding it to your ⁠~/.Rprofile⁠ startup file. Alternatively, set environment variable ⁠R_PARALLELLY_LOCALHOST_HOSTNAME=127.0.0.1⁠ in your ⁠~/.Renviron⁠ file.

If using ⁠127.0.0.1⁠ did not work around the problem, check your VPN settings and make sure it allows connections to localhost or ⁠127.0.0.1⁠.

Failing to set up remote workers

A cluster of remote workers runs R processes on external machines. These external R processes are launched over, typically, SSH to the remote machine. For this to work, each of the remote machines needs to have R installed, which preferably is of the same version as what is on the main machine. For this to work, it is required that one can SSH to the remote machines. Ideally, the SSH connections use authentication based on public-private SSH keys such that the set up of the remote workers can be fully automated (see above). If makeClusterPSOCK() fails to set up one or more remote R workers, then an informative error message is produced. There are a few reasons for failing to set up remote workers. If this happens, start by asserting that you can SSH to the remote machine and launch ‘Rscript’ by calling something like:

{local}$ ssh -l alice remote.server.org
{remote}$ Rscript --version
R scripting front-end version 4.2.2 (2022-10-31)
{remote}$ logout
{local}$

When you have confirmed the above to work, then confirm that you can achieve the same in a single command-line call;

{local}$ ssh -l alice remote.server.org Rscript --version
R scripting front-end version 4.2.2 (2022-10-31)
{local}$

The latter will assert that you have proper startup configuration also for non-interactive shell sessions on the remote machine.

If the remote machines are running on MS Windows, make sure to add argument rscript_sh = "cmd" when calling makeClusterPSOCK(), because the default is rscript_sh = "sh", which assumes that that the remote machines are Unix-like machines.

Another reason for failing to setup remote workers could be that they are running an R version that is not compatible with the version that your main R session is running. For instance, if we run R (>= 3.6.0) locally and the workers run R (< 3.5.0), we will get: ⁠Error in unserialize(node$con) : error reading from connection⁠. This is because R (>= 3.6.0) uses serialization format version 3 by default whereas R (< 3.5.0) only supports version 2. We can see the version of the R workers by adding rscript_args = c("-e", shQuote("getRversion()")) when calling makeClusterPSOCK().

For package developers

When creating a cluster object, for instance via parallel::makeCluster() or parallelly::makeClusterPSOCK(), in a package help example, in a package vignette, or in a package test, we must remember to stop the cluster at the end of all examples(*), vignettes, and unit tests. This is required in order to not leave behind stray parallel cluster workers after our main R session terminates. On Linux and macOS, the operating system often takes care of terminating the worker processes if we forget, but on MS Windows such processes will keep running in the background until they time out themselves, which takes 30 days (sic!).

⁠R CMD check --as-cran⁠ will indirectly detect these stray worker processes on MS Windows when running R (>= 4.3.0). They are detected, because they result in placeholder ⁠Rscript<hexcode>⁠ files being left behind in the temporary directory. The check NOTE to look out for (only in R (>= 4.3.0)) is:

* checking for detritus in the temp directory ... NOTE
Found the following files/directories:
  'Rscript1058267d0c10' 'Rscriptbd4267d0c10'

Those ⁠Rscript<hexcode>⁠ files are from background R worker processes, which almost always are parallel cluster:s that we forgot to stop at the end. To stop all cluster workers, use parallel::stopCluster() at the end of your examples(*), vignettes, and package tests for every cluster object that is created.

(*) Currently, examples are excluded from the detritus checks. This was validated with R-devel revision 82991 (2022-10-02).

Examples

## NOTE: Drop 'dryrun = TRUE' below in order to actually connect.  Add
## 'verbose = TRUE' if you run into problems and need to troubleshoot.

## ---------------------------------------------------------------
## Section 1. Setting up parallel workers on the local machine
## ---------------------------------------------------------------
## EXAMPLE: Two workers on the local machine
workers <- c("localhost", "localhost")
cl <- makeClusterPSOCK(workers, dryrun = TRUE, quiet = TRUE)


## EXAMPLE: Launch 124 workers on MS Windows 10, where half are
## running on CPU Group #0 and half on CPU Group #1.  
## (https://lovickconsulting.com/2021/11/18/
##  running-r-clusters-on-an-amd-threadripper-3990x-in-windows-10-2/)

## Temporarily disable CPU load protection for this example
oopts <- options(parallelly.maxWorkers.localhost = Inf)

ncores <- 124
cpu_groups <- c(0, 1)
cl <- lapply(cpu_groups, FUN = function(cpu_group) {
    parallelly::makeClusterPSOCK(ncores %/% length(cpu_groups),
      rscript = I(c(
        Sys.getenv("COMSPEC"), "/c", "start", "/B",
        "/NODE", cpu_group, "/AFFINITY", "0xFFFFFFFFFFFFFFFE",
        "*"
      )),
      dryrun = TRUE, quiet = TRUE
    )
})
## merge the two 62-node clusters into one with 124 nodes
cl <- do.call(c, cl)

## Re-enable CPU load protection
options(oopts)


## ---------------------------------------------------------------
## Section 2. Setting up parallel workers on remote machines
## ---------------------------------------------------------------
## EXAMPLE: Three remote workers
## Setup of three R workers on two remote machines are set up
workers <- c("n1.remote.org", "n2.remote.org", "n1.remote.org")
cl <- makeClusterPSOCK(workers, dryrun = TRUE, quiet = TRUE)


## EXAMPLE: Two remote workers running on MS Windows.  Because the
## remote workers are MS Windows machines, we need to use
## rscript_sh = "cmd".
workers <- c("mswin1.remote.org", "mswin2.remote.org")
cl <- makeClusterPSOCK(workers, rscript_sh = "cmd", dryrun = TRUE, quiet = TRUE)


## EXAMPLE: Local and remote workers
## Same setup when the two machines are on the local network and
## have identical software setups
cl <- makeClusterPSOCK(
  workers,
  revtunnel = FALSE, homogeneous = TRUE,
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: Three remote workers 'n1', 'n2', and 'n3' that can only be
## accessed via jumphost 'login.remote.org'
workers <- c("n1", "n2", "n1")
cl <- makeClusterPSOCK(
  workers,
  rshopts = c("-J", "login.remote.org"),
  homogeneous = FALSE,
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: Remote worker running on Linux from MS Windows machine
## Connect to remote Unix machine 'remote.server.org' on port 2200
## as user 'bob' from a MS Windows machine with PuTTY installed.
## Using the explicit special rshcmd = "<putty-plink>", will force
## makeClusterPSOCK() to search for and use the PuTTY plink software,
## preventing it from using other SSH clients on the system search PATH.
cl <- makeClusterPSOCK(
  "remote.server.org", user = "bob",
  rshcmd = "<putty-plink>",
  rshopts = c("-P", 2200, "-i", "C:/Users/bobby/.ssh/putty.ppk"),
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: Remote workers with specific setup
## Setup of remote worker with more detailed control on
## authentication and reverse SSH tunneling
cl <- makeClusterPSOCK(
  "remote.server.org", user = "johnny",
  ## Manual configuration of reverse SSH tunneling
  revtunnel = FALSE,
  rshopts = c("-v", "-R 11000:gateway:11942"),
  master = "gateway", port = 11942,
  ## Run Rscript nicely and skip any startup scripts
  rscript = c("nice", "/path/to/Rscript"),
  rscript_args = c("--no-init-file"),
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: Remote worker running on Linux from RStudio on MS Windows
## Connect to remote Unix machine 'remote.server.org' on port 2200
## as user 'bob' from a MS Windows machine via RStudio's SSH client.
## Using the explicit special rshcmd = "<rstudio-ssh>", will force
## makeClusterPSOCK() to use the SSH client that comes with RStudio,
## preventing it from using other SSH clients on the system search PATH.
cl <- makeClusterPSOCK(
  "remote.server.org:2200", user = "bob", rshcmd = "<rstudio-ssh>",
  dryrun = TRUE, quiet = TRUE
)


## ---------------------------------------------------------------
## Section 3. Setting up parallel workers on HPC cluster
## ---------------------------------------------------------------
## EXAMPLE: 'Grid Engine' is a high-performance compute (HPC) job
## scheduler where one can request compute resources on multiple nodes,
## each running multiple cores. Examples of Grid Engine schedulers are
## Oracle Grid Engine (formerly Sun Grid Engine), Univa Grid Engine,
## and Son of Grid Engine - all commonly referred to as SGE schedulers.
## Each SGE cluster may have its own configuration with their own way
## of requesting parallel slots. Here are a few examples:
##
##   ## Request 18 slots on a single host
##   qsub -pe smp 18 script.sh
##
##   ## Request 18 slots on one or more hosts
##   qsub -pe mpi 18 script.sh
##
## This will launch the job script 'script.sh' on one host, while have
## reserved in total 18 slots (CPU cores) on this host and possible
## other hosts.
##
## This example shows how to use the SGE command 'qrsh' to launch
## 18 parallel workers from R, which is assumed to have been launched
## by 'script.sh'.
cl <- makeClusterPSOCK(
  availableWorkers(),
  rshcmd = "qrsh", rshopts = c("-inherit", "-nostdin", "-V"),
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: The 'Fujitsu Technical Computing Suite' is a high-performance
## compute (HPC) job scheduler where one can request compute resources on
## multiple nodes, each running multiple cores.  For example,
##
##   pjsub -L vnode=3 -L vnode-core=18 script.sh
##
## reserves 18 cores on three nodes. The job script runs on the first
## with enviroment variables set to infer the other nodes, resulting in
## availableWorkers() to return 3 * 18 workers. When the HPC environment
## does not support SSH between compute nodes, one can use the 'pjrsh'
## command to launch the parallel workers.
cl <- makeClusterPSOCK(
  availableWorkers(),
  rshcmd = "pjrsh",
  dryrun = TRUE, quiet = TRUE
)



## ---------------------------------------------------------------
## Section 4. Setting up remote parallel workers in the cloud
## ---------------------------------------------------------------
## EXAMPLE: Remote worker running on AWS
## Launching worker on Amazon AWS EC2 running one of the
## Amazon Machine Images (AMI) provided by RStudio
## (https://www.louisaslett.com/RStudio_AMI/)
public_ip <- "1.2.3.4"
ssh_private_key_file <- "~/.ssh/my-private-aws-key.pem"
cl <- makeClusterPSOCK(
  ## Public IP number of EC2 instance
  public_ip,
  ## User name (always 'ubuntu')
  user = "ubuntu",
  ## Use private SSH key registered with AWS
  rshopts = c(
    "-o", "StrictHostKeyChecking=no",
    "-o", "IdentitiesOnly=yes",
    "-i", ssh_private_key_file
  ),
  ## Set up .libPaths() for the 'ubuntu' user
  ## and then install the future package
  rscript_startup = quote(local({
    p <- Sys.getenv("R_LIBS_USER")
    dir.create(p, recursive = TRUE, showWarnings = FALSE)
    .libPaths(p)
    install.packages("future")
  })),
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: Remote worker running on GCE
## Launching worker on Google Cloud Engine (GCE) running a
## container based VM (with a #cloud-config specification)
public_ip <- "1.2.3.4"
user <- "johnny"
ssh_private_key_file <- "~/.ssh/google_compute_engine"
cl <- makeClusterPSOCK(
  ## Public IP number of GCE instance
  public_ip,
  ## User name (== SSH key label (sic!))
  user = user,
  ## Use private SSH key registered with GCE
  rshopts = c(
    "-o", "StrictHostKeyChecking=no",
    "-o", "IdentitiesOnly=yes",
    "-i", ssh_private_key_file
  ),
  ## Launch Rscript inside Docker container
  rscript = c(
    "docker", "run", "--net=host", "rocker/r-parallel",
    "Rscript"
  ),
  dryrun = TRUE, quiet = TRUE
)



## ---------------------------------------------------------------
## Section 5. Parallel workers running locally inside virtual
## machines, Linux containers, etc.
## ---------------------------------------------------------------
## EXAMPLE: Two workers running in Docker on the local machine
## Setup of 2 Docker workers running rocker/r-parallel
cl <- makeClusterPSOCK(
  rep("localhost", times = 2L),
  ## Launch Rscript inside Docker container
  rscript = c(
    "docker", "run", "--net=host", "rocker/r-parallel",
    "Rscript"
  ),
  ## IMPORTANT: Because Docker runs inside a virtual machine (VM) on macOS
  ## and MS Windows (not Linux), when the R worker tries to connect back to
  ## the default 'localhost' it will fail, because the main R session is
  ## not running in the VM, but outside on the host.  To reach the host on
  ## macOS and MS Windows, make sure to use master = "host.docker.internal"
  master = if (.Platform$OS.type == "unix") NULL else "host.docker.internal",
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: Two workers running via Linux container 'rocker/r-parallel' from
## DockerHub on the local machine using Apptainer (formerly Singularity)
cl <- makeClusterPSOCK(
  rep("localhost", times = 2L),
  ## Launch Rscript inside Linux container
  rscript = c(
    "apptainer", "exec", "docker://rocker/r-parallel",
    "Rscript"
  ),
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: One worker running in udocker on the local machine
## Setup of a single udocker.py worker running rocker/r-parallel
cl <- makeClusterPSOCK(
  "localhost",
  ## Launch Rscript inside Docker container (using udocker)
  rscript = c(
    "udocker.py", "run", "rocker/r-parallel",
    "Rscript"
  ), 
  ## Manually launch parallel workers
  ## (need double shQuote():s because udocker.py drops one level)
  rscript_args = c(
    "-e", shQuote(shQuote("parallel:::.workRSOCK()"))
  ),
  dryrun = TRUE, quiet = TRUE
)


## EXAMPLE: One worker running in Wine for Linux on the local machine
## To install R for MS Windows in Wine, do something like:
##   winecfg  # In GUI, set 'Windows version' to 'Windows 10'
##   wget https://cran.r-project.org/bin/windows/base/R-4.2.3-win.exe
##   wine R-4.2.3-win.exe /SILENT
## Prevent packages from being installed to R's system library:
##   chmod ugo-w "$HOME/.wine/drive_c/Program Files/R/R-4.2.3/library/"
## Verify it works:
##   wine "C:/Program Files/R/R-4.2.3/bin/x64/Rscript.exe" --version
cl <- makeClusterPSOCK(1L,
  rscript = c(
    ## Silence Wine warnings
    "WINEDEBUG=fixme-all",
    ## Don't pass LC_* and R_LIBS* environments from host to Wine
    sprintf("%s=", grep("^(LC_|R_LIBS)", names(Sys.getenv()), value = TRUE)),
    "wine",
    "C:/Program Files/R/R-4.2.3/bin/x64/Rscript.exe"
  ),
  dryrun = TRUE, quiet = TRUE
)

Description

Clone one or more nodes

Usage

cloneNode(x, ...)

Arguments

Value

An object of class class(x).

Examples

cl <- makeClusterPSOCK(2)
print(cl)

## Terminate the second cluster node
parallel::stopCluster(cl[2])

## Show that cluster node #2 is no longer alive (wait a bit first)
Sys.sleep(1.0)
print(isNodeAlive(cl))
print(cl)

## "Restart" it
cl[2] <- cloneNode(cl[2])
print(cl)

## Check all nodes
print(isNodeAlive(cl))

Description

Find a TCP port that can be opened

Usage

freePort(ports = 1024:65535, default = "random", randomize = TRUE)

Arguments

Value

Returns an integer representing the first port among ports that can be opened. If none can be opened, then default is returned.

Description

Get a unique identifier for an R connection and check whether or not the connection is still valid.

Usage

isConnectionValid(con)

connectionId(con)

Arguments

Value

isConnectionValid() returns TRUE if the connection is still valid, otherwise FALSE. If FALSE, then character attribute reason provides an explanation why the connection is not valid.

connectionId() returns an non-negative integer, -1, or NA_integer_. For connections stdin, stdout, and stderr, 0, 1, and 2, are returned, respectively. For all other connections, an integer greater or equal to 3 based on the connection's internal pointer is returned. A connection that has been serialized, which is no longer valid, has identifier -1. Attribute raw_id returns the pointer string from which the above is inferred.

Connection Index versus Connection Identifier

R represents connections as indices using plain integers, e.g. idx <- as.integer(con). The three connections standard input ("stdin"), standard output ("stdout"), and standard error ("stderr") always exists and have indices 0, 1, and 2. Any connection opened beyond these will get index three or greater, depending on availability as given by base::showConnections(). To get the connection with a given index, use base::getConnection(). Unfortunately, this index representation of connections is non-robust, e.g. there are cases where two or more 'connection' objects can end up with the same index and if used, the written output may end up at the wrong destination and files and database might get corrupted. This can for instance happen if base::closeAllConnections() is used (*). In contrast, id <- connectionId(con) gives an identifier that is unique to that 'connection' object. This identifier is based on the internal pointer address of the object. The risk for two connections in the same R session to end up with the same pointer address is very small. Thus, in case we ended up in a situation where two connections con1 and con2 share the same index—as.integer(con1) == as.integer(con2)— they will never share the same identifier— connectionId(con1) != connectionId(con2). Here, isConnectionValid() can be used to check which one of these connections, if any, are valid.

(*) Note that there is no good reason for calling closeAllConnections() If called, there is a great risk that the files get corrupted etc. See (1) for examples and details on this problem. If you think there is a need to use it, it is much safer to restart R because that is guaranteed to give you a working R session with non-clashing connections. It might also be that closeAllConnections() is used because base::sys.save.image() is called, which might happen if R is being forced to terminate.

Connections Cannot be Serialized Or Saved

A 'connection' cannot be serialized, e.g. it cannot be saved to file to be read and used in another R session. If attempted, the connection will not be valid. This is a problem that may occur in parallel processing when passing an R object to parallel worker for further processing, e.g. the exported object may hold an internal database connection which will no longer be valid on the worker. When a connection is serialized, its internal pointer address will be invalidated (set to nil). In such cases, connectionId(con) returns -1 and isConnectionValid(con) returns FALSE.

References

1. 'BUG: A connection object may become corrupt and re-referenced to another connection (PATCH)', 2018-10-30.

2. R-devel thread PATCH: Asserting that 'connection' used has not changed + R_GetConnection2(), 2018-10-31.

See Also

See base::showConnections() for currently open connections and their indices. To get a connection by its index, use base::getConnection().

Examples

## R represents connections as plain indices
as.integer(stdin())          ## int 0
as.integer(stdout())         ## int 1
as.integer(stderr())         ## int 2

## The first three connections always exist and are always valid
isConnectionValid(stdin())   ## TRUE
connectionId(stdin())        ## 0L
isConnectionValid(stdout())  ## TRUE
connectionId(stdout())       ## 1L
isConnectionValid(stderr())  ## TRUE
connectionId(stderr())       ## 2L

## Connections cannot be serialized
con <- file(tempfile(), open = "w")
x <- list(value = 42, stderr = stderr(), con = con)
y <- unserialize(serialize(x, connection = NULL))
isConnectionValid(y$stderr)  ## TRUE
connectionId(y$stderr)       ##  2L
isConnectionValid(y$con)     ## FALSE with attribute 'reason'
connectionId(y$con)          ## -1L
close(con)

Description

Checks whether or not we are running in a forked child process

Usage

isForkedChild()

Details

Examples of setups and functions that rely on forked parallelization are parallel::makeCluster(n, type = "FORK"), parallel::mclapply(), and future::plan("multicore").

Value

(logical) Returns TRUE if the running in a forked child process, otherwise FALSE.

Description

Checks whether or not a Cluster Node Runs in a Forked Process

Usage

isForkedNode(node, ...)

Arguments

Value

(logical) Returns TRUE if the cluster node is running in a forked child process and FALSE if it does not. If it cannot be inferred, NA is returned.

Description

Checks whether or not a Cluster Node Runs on Localhost

Usage

isLocalhostNode(node, ...)

Arguments

Value

(logical) Returns TRUE if the cluster node is running on the current machine and FALSE if it runs on another machine. If it cannot be inferred, NA is returned.

Description

Check whether or not the cluster nodes are alive

Usage

isNodeAlive(x, ...)

Arguments

Details

This function works by checking whether the cluster node process is running or not. This is done by querying the system for its process ID (PID), which is registered by makeClusterPSOCK() when the node starts. If the PID is not known, the NA is returned. On Unix and macOS, the PID is queried using tools::pskill() with fallback to system("ps"). On MS Windows, system2("tasklist") is used, which may take a long time if there are a lot of processes running. For details, see the internal pid_exists() function.

Value

A logical vector of length length(x) with values FALSE, TRUE, and NA. If it can be established that the process for a cluster node is running, then TRUE is returned. If it does not run, then FALSE is returned. If neither can be inferred, or it times out, then NA is returned.

See Also

Use parallel::stopCluster() to shut down cluster nodes. If that's not sufficient, killNode() may be attempted.

Examples

cl <- makeClusterPSOCK(2)

## Check if cluster node #2 is alive
print(isNodeAlive(cl[[2]]))

## Check all nodes
print(isNodeAlive(cl))

Description

Terminate one or more cluster nodes using process signaling

Usage

killNode(x, signal = tools::SIGTERM, ...)

Arguments

Details

Note that the preferred way to terminate a cluster is via parallel::stopCluster(), because it terminates the cluster nodes by kindly asking each of them to nicely shut themselves down. Using killNode() is a much more sever approach. It abruptly terminates the underlying R process, possibly without giving the parallel worker a chance to terminate gracefully. For example, it might get terminated in the middle of writing to file.

tools::pskill() is used to send the signal to the R process hosting the parallel worker.

Value

TRUE if the signal was successfully applied, FALSE if not, and NA if signaling is not supported on the specific cluster or node. Warning: With R (< 3.5.0), NA is always returned. This is due to a bug in R (< 3.5.0), where the signaling result cannot be trusted.

Known limitations

This function works only with cluster nodes of class RichSOCKnode, which were created by makeClusterPSOCK(). It does not work when using parallel::makeCluster() and friends.

Currently, it's only possible to send signals to parallel workers, that is, cluster nodes, that run on the local machine. If attempted to use killNode() on a remote parallel workers, NA is returned and an informative warning is produced.

See Also

Use isNodeAlive() to check whether one or more cluster nodes are alive.

Examples

cl <- makeClusterPSOCK(2)
print(isNodeAlive(cl))  ## [1] TRUE TRUE

res <- killNode(cl)
print(res)

## It might take a moment before the background
## workers are shutdown after having been signaled
Sys.sleep(1.0)

print(isNodeAlive(cl))  ## [1] FALSE FALSE

Description

The makeClusterMPI() function creates an MPI cluster of R workers for parallel processing. This function utilizes makeCluster(..., type = "MPI") of the parallel package and tweaks the cluster in an attempt to avoid stopCluster() from hanging (1). WARNING: This function is very much in a beta version and should only be used if parallel::makeCluster(..., type = "MPI") fails.

Usage

makeClusterMPI(
  workers,
  ...,
  autoStop = FALSE,
  verbose = getOption2("parallelly.debug", FALSE)
)

Arguments

Details

Creating MPI clusters requires that the Rmpi and snow packages are installed.

Value

An object of class c("RichMPIcluster", "MPIcluster", "cluster") consisting of a list of "MPInode" workers.

References

1. R-sig-hpc thread Rmpi: mpi.close.Rslaves() 'hangs' on 2017-09-28.

See Also

makeClusterPSOCK() and parallel::makeCluster().

Examples

## Not run: 
if (requireNamespace("Rmpi") && requireNamespace("snow")) {
  cl <- makeClusterMPI(2, autoStop = TRUE)
  print(cl)
  y <- parLapply(cl, X = 1:3, fun = sqrt)
  print(y)
  rm(list = "cl")
}

## End(Not run)

Description

Below are the R options and environment variables that are used by the parallelly package and packages enhancing it.

WARNING: Note that the names and the default values of these options may change in future versions of the package. Please use with care until further notice.

Backward compatibility with the future package

The functions in the parallelly package originates from the future package. Because they are widely used within the future ecosystem, we need to keep them backward compatible for quite a long time, in order for all existing packages and R scripts to have time to adjust. This also goes for the R options and the environment variables used to configure these functions. All options and environment variables used here have prefixes parallelly. and R_PARALLELLY_, respectively. Because of the backward compatibility with the future package, the same settings can also be controlled by options and environment variables with prefixes future. and R_FUTURE_ until further notice, e.g. setting option future.availableCores.fallback=1 is the same as setting option parallelly.availableCores.fallback=1, and setting environment variable R_FUTURE_AVAILABLECORES_FALLBACK=1 is the same as setting R_PARALLELLY_AVAILABLECORES_FALLBACK=1.

Configuring number of parallel workers

The below R options and environment variables control the default results of availableCores() and availableWorkers().

parallelly.availableCores.logical:

(logical) The default value of argument logical as used by availableCores(), availableWorkers(), and availableCores() for querying parallel::detectCores(logical = logical). The default is TRUE just like it is for parallel::detectCores().

parallelly.availableCores.methods:

(character vector) Default lookup methods for availableCores(). (Default: c("system", "cgroups.cpuset", "cgroups.cpuquota", "cgroups2.cpu.max", "nproc", "mc.cores", "BiocParallel", "_R_CHECK_LIMIT_CORES_", "Bioconductor", "LSF", "PJM", "PBS", "SGE", "Slurm", "fallback", "custom"))

parallelly.availableCores.custom:

(function) If set and a function, then this function will be called (without arguments) by availableCores() where its value, coerced to an integer, is interpreted as a number of cores.

parallelly.availableCores.fallback:

(integer) Number of cores to use when no core-specifying settings are detected other than "system" and "nproc". This options makes it possible to set the default number of cores returned by availableCores() / availableWorkers() yet allow users and schedulers to override it. In multi-tenant environment, such as HPC clusters, it is useful to set environment variable R_PARALLELLY_AVAILABLECORES_FALLBACK to 1, which will set this option when the package is loaded.

parallelly.availableCores.system:

(integer) Number of "system" cores used instead of what is reported by availableCores(which = "system"). This option allows you to effectively override what parallel::detectCores() reports the system has.

parallelly.availableCores.min:

(integer) The minimum number of cores availableCores() is allowed to return. This can be used to force multiple cores on a single-core environment. If this is limit is applied, the names of the returned value are appended with an asterisk (*). (Default: 1L)

parallelly.availableCores.omit:

(integer) Number of cores to set aside, i.e. not to include.

parallelly.availableWorkers.methods:

(character vector) Default lookup methods for availableWorkers(). (Default: c("mc.cores", "BiocParallel", "_R_CHECK_LIMIT_CORES_", "Bioconductor", "LSF", "PJM", "PBS", "SGE", "Slurm", "custom", "cgroups.cpuset", "cgroups.cpuquota", "cgroups2.cpu.max", "nproc", "system", "fallback"))

parallelly.availableWorkers.custom:

(function) If set and a function, then this function will be called (without arguments) by availableWorkers() where its value, coerced to a character vector, is interpreted as hostnames of available workers.

Configuring forked parallel processing

The below R options and environment variables control the default result of supportsMulticore().

parallelly.fork.enable:

(logical) Enable or disable forked processing. If FALSE, multicore futures becomes sequential futures. If NA, or not set (the default), the a set of best-practices rules decide whether should be supported or not.

parallelly.supportsMulticore.disableOn:

(character vector) because the environment in which R runs is considered unstable for forked processing. If this vector contains "rstudio_console", it is disabled when running R in the RStudio Console. If this vector contains "rstudio_terminal", it is disabled when running R in the RStudio Terminal. (Default: c("rstudio_console", "rstudio_terminal"))

parallelly.supportsMulticore.unstable:

(character) Controls whether a warning should be produced or not whenever multicore processing is automatically disabled per settings in option parallelly.supportsMulticore.disableOn. If "warn" (default), then an informative warning is produces the first time 'multicore' futures are used. If "quiet", no warning is produced.

Configuring setup of parallel PSOCK clusters

The below R options and environment variables control the default results of makeClusterPSOCK() and its helper function makeNodePSOCK() that creates the individual cluster nodes.

parallelly.maxWorkers.localhost:

(two numerics) Maximum number of localhost workers, relative to availableCores(), accepted and allowed. The first element corresponds to the threshold where a warning is produced, the second where an error is produced. Thresholds may be +Inf. If only the first exist, no error is produced (defaults to c(1.0, 3.0) corresponding to a maximum 100% and 300% use).

parallelly.makeNodePSOCK.setup_strategy:

(character) If "parallel" (default), the PSOCK cluster nodes are set up concurrently. If "sequential", they are set up sequentially.

parallelly.makeNodePSOCK.validate:

(logical) If TRUE (default), after the nodes have been created, they are all validated that they work by inquiring about their session information, which is saved in attribute session_info of each node.

parallelly.makeNodePSOCK.connectTimeout:

(numeric) The maximum time (in seconds) allowed for each socket connection between the master and a worker to be established (defaults to 2*60 seconds = 2 minutes).

parallelly.makeNodePSOCK.timeout:

(numeric) The maximum time (in seconds) allowed to pass without the master and a worker communicate with each other (defaults to 302460*60 seconds = 30 days).

parallelly.makeNodePSOCK.useXDR:

(logical) If FALSE (default), the communication between master and workers, which is binary, will use small-endian (faster), otherwise big-endian ("XDR"; slower).

parallelly.makeNodePSOCK.socketOptions:

(character string) If set to another value than "NULL", then option socketOptions is set to this value on the workers during startup. See base::socketConnection() for details. (defaults to "no-delay")

parallelly.makeNodePSOCK.rshcmd:

(character vector) The command to be run on the master to launch a process on another host.

parallelly.makeNodePSOCK.rshopts:

(character vector) Addition command-line options appended to rshcmd. These arguments are only applied when connecting to non-localhost machines.

parallelly.makeNodePSOCK.tries:

(integer) The maximum number of attempts done to launch each node. Only used when setting up cluster nodes using the sequential strategy.

parallelly.makeNodePSOCK.tries.delay:

(numeric) The number of seconds to wait before trying to launch a cluster node that failed to launch previously. Only used when setting up cluster nodes using the sequential strategy.

Options for debugging

parallelly.debug:

(logical) If TRUE, extensive debug messages are generated. (Default: FALSE)

Environment variables that set R options

All of the above R ⁠parallelly.*⁠ options can be set by corresponding environment variables R_PARALLELLY_* when the parallelly package is loaded. For example, if R_PARALLELLY_MAKENODEPSOCK_SETUP_STRATEGY = "sequential", then option parallelly.makeNodePSOCK.setup_strategy is set to "sequential" (character). Similarly, if R_PARALLELLY_AVAILABLECORES_FALLBACK = "1", then option parallelly.availableCores.fallback is set to 1 (integer).

See Also

To set R options when R starts (even before the parallelly package is loaded), see the Startup help page. The startup package provides a friendly mechanism for configuring R's startup process.

Examples

# Set an R option:
options(parallelly.availableCores.fallback = 1L)

Description

Certain parallelization methods in R rely on forked processing, e.g. parallel::mclapply(), parallel::makeCluster(n, type = "FORK"), doMC::registerDoMC(), and future::plan("multicore"). Process forking is done by the operating system and support for it in R is restricted to Unix-like operating systems such as Linux, Solaris, and macOS. R running on Microsoft Windows does not support forked processing. In R, forked processing is often referred to as "multicore" processing, which stems from the 'mc' of the mclapply() family of functions, which originally was in a package named multicore which later was incorporated into the parallel package. This function checks whether or not forked (aka "multicore") processing is supported in the current R session.

Usage

supportsMulticore(...)

Arguments

Value

TRUE if forked processing is supported and not disabled, otherwise FALSE.

Support for process forking

While R supports forked processing on Unix-like operating system such as Linux and macOS, it does not on the Microsoft Windows operating system.

For some R environments it is considered unstable to perform parallel processing based on forking. This is for example the case when using RStudio, cf. RStudio Inc. recommends against using forked processing when running R from within the RStudio software. This function detects when running in such an environment and returns FALSE, despite the underlying operating system supports forked processing. A warning will also be produced informing the user about this the first time time this function is called in an R session. This warning can be disabled by setting R option parallelly.supportsMulticore.unstable, or environment variable R_PARALLELLY_SUPPORTSMULTICORE_UNSTABLE to "quiet".

Enable or disable forked processing

It is possible to disable forked processing for futures by setting R option parallelly.fork.enable to FALSE. Alternatively, one can set environment variable R_PARALLELLY_FORK_ENABLE to false. Analogously, it is possible to override disabled forking by setting one of these to TRUE.

Examples

## Check whether or not forked processing is supported
supportsMulticore()