Version:	1.58.0
Title:	Unified Parallel and Distributed Processing in R for Everyone
Depends:	R (≥ 3.2.0)
Imports:	digest, globals (≥ 0.18.0), listenv (≥ 0.8.0), parallel, parallelly (≥ 1.44.0), utils
Suggests:	methods, RhpcBLASctl, R.rsp, markdown
VignetteBuilder:	R.rsp
Description:	The purpose of this package is to provide a lightweight and unified Future API for sequential and parallel processing of R expression via futures. The simplest way to evaluate an expression in parallel is to use 'x %<-% { expression }' with 'plan(multisession)'. This package implements sequential, multicore, multisession, and cluster futures. With these, R expressions can be evaluated on the local machine, in parallel a set of local machines, or distributed on a mix of local and remote machines. Extensions to this package implement additional backends for processing futures via compute cluster schedulers, etc. Because of its unified API, there is no need to modify any code in order switch from sequential on the local machine to, say, distributed processing on a remote compute cluster. Another strength of this package is that global variables and functions are automatically identified and exported as needed, making it straightforward to tweak existing code to make use of futures.
License:	LGPL-2.1 | LGPL-3 [expanded from: LGPL (≥ 2.1)]
LazyLoad:	TRUE
ByteCompile:	TRUE
URL:	https://future.futureverse.org, https://github.com/futureverse/future
BugReports:	https://github.com/futureverse/future/issues
Language:	en-US
Encoding:	UTF-8
RoxygenNote:	7.3.2
NeedsCompilation:	no
Packaged:	2025-06-05 01:03:35 UTC; henrik
Author:	Henrik Bengtsson [aut, cre, cph]
Maintainer:	Henrik Bengtsson <henrikb@braju.com>
Repository:	CRAN
Date/Publication:	2025-06-05 05:30:06 UTC

Gets the length of an object without dispatching

Description

Gets the length of an object without dispatching

Usage

.length(x)

Arguments

Details

This function returns length(unclass(x)), but tries to avoid calling unclass(x) unless necessary.

Value

A non-negative integer.

See Also

.subset() and .subset2().

A future represents a value that will be available at some point in the future

Description

A future is an abstraction for a value that may available at some point in the future. A future can either be unresolved or resolved, a state which can be checked with resolved(). As long as it is unresolved, the value is not available. As soon as it is resolved, the value is available via value().

Usage

Future(
  expr = NULL,
  envir = parent.frame(),
  substitute = TRUE,
  stdout = TRUE,
  conditions = "condition",
  globals = list(),
  packages = NULL,
  seed = FALSE,
  lazy = FALSE,
  gc = FALSE,
  earlySignal = FALSE,
  label = NULL,
  ...
)

Arguments

Details

A Future object is itself an environment.

Value

Future() returns an object of class Future.

See Also

One function that creates a Future is future(). It returns a Future that evaluates an R expression in the future. An alternative approach is to use the %<-% infix assignment operator, which creates a future from the right-hand-side (RHS) R expression and assigns its future value to a variable as a promise.

Configure a backend that controls how and where futures are evaluated

Description

This functionality is only for developers who wish to implement their own future backend. End-users and package developers use futureverse, does not need to know about these functions.

If you are looking for available future backends to choose from, please see the 'A Future for R: Available Future Backends' vignette and https://www.futureverse.org/backends.html.

Usage

FutureBackend(
  ...,
  earlySignal = FALSE,
  gc = FALSE,
  maxSizeOfObjects = getOption("future.globals.maxSize", +Inf),
  interrupts = TRUE,
  hooks = FALSE
)

launchFuture(backend, future, ...)

listFutures(backend, ...)

interruptFuture(backend, future, ...)

validateFutureGlobals(backend, future, ...)

stopWorkers(backend, ...)

MultiprocessFutureBackend(
  ...,
  wait.timeout = getOption("future.wait.timeout", 24 * 60 * 60),
  wait.interval = getOption("future.wait.interval", 0.01),
  wait.alpha = getOption("future.wait.alpha", 1.01)
)

ClusterFutureBackend(
  workers = availableWorkers(constraints = "connections"),
  gc = TRUE,
  earlySignal = TRUE,
  interrupts = FALSE,
  persistent = FALSE,
  ...
)

MulticoreFutureBackend(
  workers = availableCores(constraints = "multicore"),
  maxSizeOfObjects = +Inf,
  ...
)

SequentialFutureBackend(..., maxSizeOfObjects = +Inf)

MultisessionFutureBackend(
  workers = availableCores(),
  rscript_libs = .libPaths(),
  interrupts = TRUE,
  gc = FALSE,
  earlySignal = FALSE,
  ...
)

Arguments

Details

The ClusterFutureBackend is selected by plan(cluster, workers = workers).

The MulticoreFutureBackend backend is selected by plan(multicore, workers = workers).

The SequentialFutureBackend is selected by plan(sequential).

The MultisessionFutureBackend backend is selected by plan(multisession, workers = workers).

Value

FutureBackend() returns a FutureBackend object, which inherits an environment. Specific future backends are defined by subclasses implementing the FutureBackend API.

launchFuture() returns the launched Future object.

interruptFuture() returns the interrupted Future object, if supported, other the unmodified future.

stopWorkers() returns TRUE if the workers were shut down, otherwise FALSE.

The FutureBackend API

The FutureBackend class specifies FutureBackend API, that all backends must implement and comply to. Specifically,

See Also

For alternative future backends, see the 'A Future for R: Available Future Backends' vignette and https://www.futureverse.org/backends.html.

For alternative future backends, see the 'A Future for R: Available Future Backends' vignette and https://www.futureverse.org/backends.html.

A condition (message, warning, or error) that occurred while orchestrating a future

Description

While orchestrating (creating, launching, querying, collection) futures, unexpected run-time errors (and other types of conditions) may occur. Such conditions are coerced to a corresponding FutureCondition class to help distinguish them from conditions that occur due to the evaluation of the future.

Usage

FutureCondition(message, call = NULL, uuid = future[["uuid"]], future = NULL)

FutureMessage(message, call = NULL, uuid = future[["uuid"]], future = NULL)

FutureWarning(message, call = NULL, uuid = future[["uuid"]], future = NULL)

FutureError(message, call = NULL, uuid = future[["uuid"]], future = NULL)

RngFutureCondition(
  message = NULL,
  call = NULL,
  uuid = future[["uuid"]],
  future = NULL
)

RngFutureWarning(...)

RngFutureError(...)

UnexpectedFutureResultError(future, hint = NULL)

GlobalEnvMisuseFutureCondition(
  message = NULL,
  call = NULL,
  differences = NULL,
  uuid = future[["uuid"]],
  future = NULL
)

GlobalEnvMisuseFutureWarning(...)

GlobalEnvMisuseFutureError(...)

ConnectionMisuseFutureCondition(
  message = NULL,
  call = NULL,
  differences = NULL,
  uuid = future[["uuid"]],
  future = NULL
)

ConnectionMisuseFutureWarning(...)

ConnectionMisuseFutureError(...)

DeviceMisuseFutureCondition(
  message = NULL,
  call = NULL,
  differences = NULL,
  uuid = future[["uuid"]],
  future = NULL
)

DeviceMisuseFutureWarning(...)

DeviceMisuseFutureError(...)

DefaultDeviceMisuseFutureCondition(
  message = NULL,
  incidents = NULL,
  call = NULL,
  uuid = future[["uuid"]],
  future = NULL
)

DefaultDeviceMisuseFutureWarning(...)

DefaultDeviceMisuseFutureError(...)

FutureLaunchError(..., future = NULL)

FutureInterruptError(..., future = NULL)

FutureCanceledError(..., future = NULL)

FutureDroppedError(..., future = NULL)

FutureJournalCondition(
  message,
  journal,
  call = NULL,
  uuid = future[["uuid"]],
  future = NULL
)

Arguments

Value

An object of class FutureCondition which inherits from class condition and FutureMessage, FutureWarning, and FutureError all inherits from FutureCondition. Moreover, a FutureError inherits from error, a FutureWarning from warning, and a FutureMessage from message.

A representation of a set of globals used with futures

Description

A representation of a set of globals used with futures

Usage

FutureGlobals(object = list(), resolved = FALSE, total_size = NA_real_, ...)

Arguments

Details

This class extends the Globals class by adding attributes resolved and total_size.

Value

An object of class FutureGlobals.

Results from resolving a future

Description

Results from resolving a future

Usage

FutureResult(
  value = NULL,
  visible = TRUE,
  stdout = NULL,
  conditions = NULL,
  rng = FALSE,
  ...,
  uuid = NULL,
  started = .POSIXct(NA_real_),
  finished = Sys.time(),
  version = "1.8"
)

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

## S3 method for class 'FutureResult'
print(x, ...)

Arguments

Details

This function is only part of the backend Future API. This function is not part of the frontend Future API.

Value

An object of class FutureResult.

Note to developers

The FutureResult structure is under development and may change at anytime, e.g. elements may be renamed or removed. Because of this, please avoid accessing the elements directly in code. Feel free to reach out if you need to do so in your code.

A multicore future is a future whose value will be resolved asynchronously in a parallel process

Description

A multicore future is a future whose value will be resolved asynchronously in a parallel process

Usage

## S3 method for class 'MulticoreFuture'
resolved(x, run = TRUE, timeout = NULL, ...)

Value

MulticoreFuture() returns an object of class MulticoreFuture.

Usage

To use 'multicore' futures, use plan(multicore, ...), cf. multicore.

A multiprocess future is a future whose value will be resolved asynchronously in a parallel process

Description

A multiprocess future is a future whose value will be resolved asynchronously in a parallel process

Usage

MultiprocessFuture(expr = NULL, substitute = TRUE, envir = parent.frame(), ...)

Arguments

Value

MultiprocessFuture() returns an object of class MultiprocessFuture.

An uniprocess future is a future whose value will be resolved synchronously in the current process

Description

An uniprocess future is a future whose value will be resolved synchronously in the current process

Usage

UniprocessFuture(expr = NULL, substitute = TRUE, envir = parent.frame(), ...)

Arguments

Value

UniprocessFuture() returns an object of class UniprocessFuture.

Back trace the expressions evaluated when an error was caught

Description

Back trace the expressions evaluated when an error was caught

Usage

backtrace(future, envir = parent.frame(), ...)

Arguments

Value

A list with the future's call stack that led up to the error.

Examples

my_log <- function(x) log(x)
foo <- function(...) my_log(...)

f <- future({ foo("a") })
res <- tryCatch({
  v <- value(f)
}, error = function(ex) {
  t <- backtrace(f)
  print(t)
})

Cancel a future

Description

Cancels futures, with the option to interrupt running ones.

Usage

cancel(x, interrupt = TRUE, ...)

Arguments

Value

cancel() returns (invisibly) the canceled Futures after flagging them as "canceled" and possibly interrupting them as well.

Canceling a lazy or a finished future has no effect.

See Also

A canceled future can be reset() to a lazy, vanilla future such that it can be relaunched, possible on another future backend.

Examples

## Set up two parallel workers
plan(multisession, workers = 2)

## Launch two long running future
fs <- lapply(c(1, 2), function(duration) {
  future({
    Sys.sleep(duration)
    42
  })
})

## Wait until at least one of the futures is resolved
while (!any(resolved(fs))) Sys.sleep(0.1)

## Cancel the future that is not yet resolved
r <- resolved(fs)
cancel(fs[!r])

## Get the value of the resolved future
f <- fs[r]
v <- value(f)
message("Result: ", v)

## The value of the canceled future is an error
try(v <- value(fs[!r]))

## Shut down parallel workers
plan(sequential)

Create a cluster future whose value will be resolved asynchronously in a parallel process

Description

WARNING: This function must never be called. It may only be used with plan()

Usage

cluster(
  ...,
  workers = availableWorkers(constraints = "connections"),
  gc = FALSE,
  earlySignal = FALSE,
  persistent = FALSE,
  envir = parent.frame()
)

Arguments

Details

A cluster future is a future that uses cluster evaluation, which means that its value is computed and resolved in parallel in another process.

This function is must not be called directly. Instead, the typical usages are:

# Evaluate futures via a single background R process on the local machine
plan(cluster, workers = I(1))

# Evaluate futures via two background R processes on the local machine
plan(cluster, workers = 2)

# Evaluate futures via a single R process on another machine on on the
# local area network (LAN)
plan(cluster, workers = "raspberry-pi")

# Evaluate futures via a single R process running on a remote machine
plan(cluster, workers = "pi.example.org")

# Evaluate futures via four R processes, one running on the local machine,
# two running on LAN machine 'n1' and one on a remote machine
plan(cluster, workers = c("localhost", "n1", "n1", "pi.example.org"))

Value

A ClusterFuture.

See Also

For alternative future backends, see the 'A Future for R: Available Future Backends' vignette and https://www.futureverse.org/backends.html.

Examples

## Use cluster futures
cl <- parallel::makeCluster(2, timeout = 60)
plan(cluster, workers = cl)

## A global variable
a <- 0

## Create future (explicitly)
f <- future({
  b <- 3
  c <- 2
  a * b * c
})

## A cluster future is evaluated in a separate process.
## Regardless, changing the value of a global variable will
## not affect the result of the future.
a <- 7
print(a)

v <- value(f)
print(v)
stopifnot(v == 0)

## CLEANUP
parallel::stopCluster(cl)

Export globals to the sticky-globals environment of the cluster nodes

Description

Export globals to the sticky-globals environment of the cluster nodes

Usage

clusterExportSticky(cl, globals)

Arguments

Details

This requires that the future package is installed on the cluster nodes.

Value

(invisible; cluster) The cluster object.

Get the first or all references of an R object

Description

Get the first or all references of an R object

Assert that there are no references among the identified globals

Usage

find_references(x, first_only = FALSE)

assert_no_references(
  x,
  action = c("error", "warning", "message", "string"),
  source = c("globals", "value")
)

Arguments

Value

find_references() returns a list of zero or more references identified.

If a reference is detected, an informative error, warning, message, or a character string is produced, otherwise NULL is returned invisibly.

Create a future

Description

Creates a future that evaluates an R expression or a future that calls an R function with a set of arguments. How, when, and where these futures are evaluated can be configured using plan() such that it is evaluated in parallel on, for instance, the current machine, on a remote machine, or via a job queue on a compute cluster. Importantly, any R code using futures remains the same regardless on these settings and there is no need to modify the code when switching from, say, sequential to parallel processing.

Usage

future(
  expr,
  envir = parent.frame(),
  substitute = TRUE,
  lazy = FALSE,
  seed = FALSE,
  globals = TRUE,
  packages = NULL,
  stdout = TRUE,
  conditions = "condition",
  label = NULL,
  gc = FALSE,
  earlySignal = FALSE,
  ...
)

futureCall(
  FUN,
  args = list(),
  envir = parent.frame(),
  lazy = FALSE,
  seed = FALSE,
  globals = TRUE,
  packages = NULL,
  stdout = TRUE,
  conditions = "condition",
  earlySignal = FALSE,
  label = NULL,
  gc = FALSE,
  ...
)

minifuture(
  expr,
  substitute = TRUE,
  globals = NULL,
  packages = NULL,
  stdout = NA,
  conditions = NULL,
  seed = NULL,
  ...,
  envir = parent.frame()
)

Arguments

Details

The state of a future is either unresolved or resolved. The value of a future can be retrieved using v <- value(f). Querying the value of a non-resolved future will block the call until the future is resolved. It is possible to check whether a future is resolved or not without blocking by using resolved(f). It is possible to cancel() a future that is being resolved. Failed, canceled, and interrupted futures can be reset() to a lazy, vanilla future that can be relaunched.

The futureCall() function works analogously to do.call(), which calls a function with a set of arguments. The difference is that do.call() returns the value of the call whereas futureCall() returns a future.

Value

future() returns Future that evaluates expression expr.

futureCall() returns a Future that calls function FUN with arguments args.

minifuture(expr) creates a future with minimal overhead, by disabling user-friendly behaviors, e.g. automatic identification of global variables and packages needed, and relaying of output. Unless you have good reasons for using this function, please use future() instead. This function exists mainly for the purpose of profiling and identifying which automatic features of future() introduce extra overhead.

Eager or lazy evaluation

By default, a future is resolved using eager evaluation (lazy = FALSE). This means that the expression starts to be evaluated as soon as the future is created.

As an alternative, the future can be resolved using lazy evaluation (lazy = TRUE). This means that the expression will only be evaluated when the value of the future is requested. Note that this means that the expression may not be evaluated at all - it is guaranteed to be evaluated if the value is requested.

Globals used by future expressions

Global objects (short globals) are objects (e.g. variables and functions) that are needed in order for the future expression to be evaluated while not being local objects that are defined by the future expression. For example, in

  a <- 42
  f <- future({ b <- 2; a * b })

variable a is a global of future assignment f whereas b is a local variable. In order for the future to be resolved successfully (and correctly), all globals need to be gathered when the future is created such that they are available whenever and wherever the future is resolved.

The default behavior (globals = TRUE), is that globals are automatically identified and gathered. More precisely, globals are identified via code inspection of the future expression expr and their values are retrieved with environment envir as the starting point (basically via get(global, envir = envir, inherits = TRUE)). In most cases, such automatic collection of globals is sufficient and less tedious and error prone than if they are manually specified.

However, for full control, it is also possible to explicitly specify exactly which the globals are by providing their names as a character vector. In the above example, we could use

  a <- 42
  f <- future({ b <- 2; a * b }, globals = "a")

Yet another alternative is to explicitly specify also their values using a named list as in

  a <- 42
  f <- future({ b <- 2; a * b }, globals = list(a = a))

or

  f <- future({ b <- 2; a * b }, globals = list(a = 42))

Specifying globals explicitly avoids the overhead added from automatically identifying the globals and gathering their values. Furthermore, if we know that the future expression does not make use of any global variables, we can disable the automatic search for globals by using

  f <- future({ a <- 42; b <- 2; a * b }, globals = FALSE)

Future expressions often make use of functions from one or more packages. As long as these functions are part of the set of globals, the future package will make sure that those packages are attached when the future is resolved. Because there is no need for such globals to be frozen or exported, the future package will not export them, which reduces the amount of transferred objects. For example, in

  x <- rnorm(1000)
  f <- future({ median(x) })

variable x and median() are globals, but only x is exported whereas median(), which is part of the stats package, is not exported. Instead it is made sure that the stats package is on the search path when the future expression is evaluated. Effectively, the above becomes

  x <- rnorm(1000)
  f <- future({
    library(stats)
    median(x)
  })

To manually specify this, one can either do

  x <- rnorm(1000)
  f <- future({
    median(x)
  }, globals = list(x = x, median = stats::median)

or

  x <- rnorm(1000)
  f <- future({
    library(stats)
    median(x)
  }, globals = list(x = x))

Both are effectively the same.

Although rarely needed, a combination of automatic identification and manual specification of globals is supported via attributes add (to add false negatives) and ignore (to ignore false positives) on value TRUE. For example, with globals = structure(TRUE, ignore = "b", add = "a") any globals automatically identified, except b, will be used, in addition to global a.

Author(s)

The future logo was designed by Dan LaBar and tweaked by Henrik Bengtsson.

See Also

How, when and where futures are resolved is given by the future backend, which can be set by the end user using the plan() function.

Examples

## Evaluate futures in parallel
plan(multisession)

## Data
x <- rnorm(100)
y <- 2 * x + 0.2 + rnorm(100)
w <- 1 + x ^ 2


## EXAMPLE: Regular assignments (evaluated sequentially)
fitA <- lm(y ~ x, weights = w)      ## with offset
fitB <- lm(y ~ x - 1, weights = w)  ## without offset
fitC <- {
  w <- 1 + abs(x)  ## Different weights
  lm(y ~ x, weights = w)
}
print(fitA)
print(fitB)
print(fitC)


## EXAMPLE: Future assignments (evaluated in parallel)
fitA %<-% lm(y ~ x, weights = w)      ## with offset
fitB %<-% lm(y ~ x - 1, weights = w)  ## without offset
fitC %<-% {
  w <- 1 + abs(x)
  lm(y ~ x, weights = w)
}
print(fitA)
print(fitB)
print(fitC)


## EXAMPLE: Explicitly create futures (evaluated in parallel)
## and retrieve their values
fA <- future( lm(y ~ x, weights = w) )
fB <- future( lm(y ~ x - 1, weights = w) )
fC <- future({
  w <- 1 + abs(x)
  lm(y ~ x, weights = w)
})
fitA <- value(fA)
fitB <- value(fB)
fitC <- value(fC)
print(fitA)
print(fitB)
print(fitC)


## EXAMPLE: futureCall() and do.call()
x <- 1:100
y0 <- do.call(sum, args = list(x))
print(y0)

f1 <- futureCall(sum, args = list(x))
y1 <- value(f1)
print(y1)

Create a future assignment

Description

x %<-% value (also known as a "future assignment") and futureAssign("x", value) create a Future that evaluates the expression (value) and binds it to variable x (as a promise). The expression is evaluated in parallel in the background. Later on, when x is first queried, the value of future is automatically retrieved as it were a regular variable and x is materialized as a regular value.

Usage

futureAssign(
  x,
  value,
  envir = parent.frame(),
  substitute = TRUE,
  lazy = FALSE,
  seed = FALSE,
  globals = TRUE,
  packages = NULL,
  stdout = TRUE,
  conditions = "condition",
  earlySignal = FALSE,
  label = NULL,
  gc = FALSE,
  ...,
  assign.env = envir
)

x %<-% value

fassignment %globals% globals
fassignment %packages% packages

fassignment %seed% seed

fassignment %stdout% capture

fassignment %conditions% capture

fassignment %lazy% lazy

fassignment %label% label

fassignment %plan% strategy

fassignment %tweak% tweaks

Arguments

Details

For a future created via a future assignment, x %<-% value or futureAssign("x", value), the value is bound to a promise, which when queried will internally call value() on the future and which will then be resolved into a regular variable bound to that value. For example, with future assignment x %<-% value, the first time variable x is queried the call blocks if, and only if, the future is not yet resolved. As soon as it is resolved, and any succeeding queries, querying x will immediately give the value.

The future assignment construct x %<-% value is not a formal assignment per se, but a binary infix operator on objects x and expression value. However, by using non-standard evaluation, this constructs can emulate an assignment operator similar to x <- value. Due to R's precedence rules of operators, future expressions often need to be explicitly bracketed, e.g. x %<-% { a + b }.

Value

futureAssign() and x %<-% expr returns the Future invisibly, e.g. f <- futureAssign("x", expr) and f <- (x %<-% expr).

Adjust future arguments of a future assignment

future() and futureAssign() take several arguments that can be used to explicitly specify what global variables and packages the future should use. They can also be used to override default behaviors of the future, e.g. whether output should be relayed or not. When using a future assignment, these arguments can be specified via corresponding assignment expression. For example, x %<-% { rnorm(10) } %seed% TRUE corresponds to futureAssign("x", { rnorm(10) }, seed = TRUE). Here are a several examples.

To explicitly specify variables and functions that a future assignment should use, use ⁠%globals%⁠. To explicitly specify which packages need to be attached for the evaluate to success, use ⁠%packages%⁠. For example,

> x <- rnorm(1000)
> y %<-% { median(x) } %globals% list(x = x) %packages% "stats"
> y
[1] -0.03956372

The median() function is part of the 'stats' package.

To declare that you will generate random numbers, use ⁠%seed%⁠, e.g.

> x %<-% { rnorm(3) } %seed% TRUE
> x
[1] -0.2590562 -1.2262495  0.8858702

To disable relaying of standard output (e.g. print(), cat(), and str()), while keeping relaying of conditions (e.g. message() and

> x %<-% { cat("Hello\n"); message("Hi there"); 42 } %stdout% FALSE
> y <- 13
> z <- x + y
Hi there
> z
[1] 55

To disable relaying of conditions, use ⁠%conditions%⁠, e.g.

> x %<-% { cat("Hello\n"); message("Hi there"); 42 } %conditions% character(0)
> y <- 13
> z <- x + y
Hello
> z
[1] 55

> x %<-% { print(1:10); message("Hello"); 42 } %stdout% FALSE
> y <- 13
> z <- x + y
Hello
> z
[1] 55

To create a future without launching in such that it will only be processed if the value is really needed, use ⁠%lazy%⁠, e.g.

> x %<-% { Sys.sleep(5); 42 } %lazy% TRUE
> y <- sum(1:10)
> system.time(z <- x + y)
  user  system elapsed 
  0.004   0.000   5.008
> z
[1] 97

Error handling

Because future assignments are promises, errors produced by the the future expression will not be signaled until the value of the future is requested. For example, if you create a future assignment that produce an error, you will not be affected by the error until you "touch" the future-assignment variable. For example,

> x %<-% { stop("boom") }
> y <- sum(1:10)
> z <- x + y
Error in eval(quote({ : boom

Use alternative future backend for future assignment

Futures are evaluated on the future backend that the user has specified by plan(). With regular futures, we can temporarily use another future backend by wrapping our code in ⁠with(plan(...), { ... }]⁠, or temporarily inside a function using with(plan(...), local = TRUE). To achieve the same for a specific future assignment, use ⁠%plan%⁠, e.g.

> plan(multisession)
> x %<-% { 42 }
> y %<-% { 13 } %plan% sequential
> z <- x + y
> z
[1] 55

Here x is resolved in the background via the multisession backend, whereas y is resolved sequentially in the main R session.

Getting the future object of a future assignment

The underlying Future of a future variable x can be retrieved without blocking using f <- futureOf(x), e.g.

> x %<-% { stop("boom") }
> f_x <- futureOf(x)
> resolved(f_x)
[1] TRUE
> x
Error in eval(quote({ : boom
> value(f_x)
Error in eval(quote({ : boom

Technically, both the future and the variable (promise) are assigned at the same time to environment assign.env where the name of the future is ⁠.future_<name>⁠.

Get the future of a future variable

Description

Get the future of a future variable that has been created directly or indirectly via future().

Usage

futureOf(
  var = NULL,
  envir = parent.frame(),
  mustExist = TRUE,
  default = NA,
  drop = FALSE
)

Arguments

Value

A Future (or default). If var is NULL, then a named list of Future:s are returned.

Examples

a %<-% { 1 }

f <- futureOf(a)
print(f)

b %<-% { 2 }

f <- futureOf(b)
print(f)

## All futures
fs <- futureOf()
print(fs)


## Futures part of environment
env <- new.env()
env$c %<-% { 3 }

f <- futureOf(env$c)
print(f)

f2 <- futureOf(c, envir = env)
print(f2)

f3 <- futureOf("c", envir = env)
print(f3)

fs <- futureOf(envir = env)
print(fs)

Get future-specific session information and validate current backend

Description

Get future-specific session information and validate current backend

Usage

futureSessionInfo(test = TRUE, anonymize = TRUE)

Arguments

Value

Nothing.

Examples

plan(multisession, workers = 2)
futureSessionInfo()
plan(sequential)

Get all futures in a container

Description

Gets all futures in an environment, a list, or a list environment and returns an object of the same class (and dimensions). Non-future elements are returned as is.

Usage

futures(x, ...)

Arguments

Details

This function is useful for retrieve futures that were created via future assignments (⁠%<-%⁠) and therefore stored as promises. This function turns such promises into standard Future objects.

Value

An object of same type as x and with the same names and/or dimensions, if set.

Inject code for the next type of future to use for nested futures

Description

Inject code for the next type of future to use for nested futures

Usage

getExpression(future, ...)

Arguments

Details

If there is no future backend specified after this one, the default is to use sequential futures. This conservative approach protects against spawning off recursive futures by mistake, especially multicore and multisession ones. The default will also set options(mc.cores = 1L) (*) so that no parallel R processes are spawned off by functions such as parallel::mclapply() and friends.

Currently it is not possible to specify what type of nested futures to be used, meaning the above default will always be used. See Issue #37 for plans on adding support for custom nested future types.

(*) Ideally we would set mc.cores = 0 but that will unfortunately cause mclapply() and friends to generate an error saying "'mc.cores' must be >= 1". Ideally those functions should fall back to using the non-multicore alternative in this case, e.g. mclapply(...) => lapply(...). See https://github.com/HenrikBengtsson/Wishlist-for-R/issues/7 for a discussion on this.

Value

A future expression with code injected to set what type of future to use for nested futures, iff any.

Retrieves global variables of an expression and their associated packages

Description

Retrieves global variables of an expression and their associated packages

Usage

getGlobalsAndPackages(
  expr,
  envir = parent.frame(),
  tweak = tweakExpression,
  globals = TRUE,
  locals = getOption("future.globals.globalsOf.locals", TRUE),
  resolve = getOption("future.globals.resolve"),
  persistent = FALSE,
  maxSize = getOption("future.globals.maxSize", 500 * 1024^2),
  onReference = getOption("future.globals.onReference", "ignore"),
  ...
)

Arguments

Value

A named list with elements expr (the tweaked expression), globals (a named list of class FutureGlobals) and packages (a character string).

See Also

Internally, globalsOf() is used to identify globals and associated packages from the expression.

Create a Future Cluster of Stateless Workers for Parallel Processing

Description

WARNING: Please note that this sets up a stateless set of cluster nodes, which means that clusterEvalQ(cl, { a <- 3.14 }) will have no effect. Consider this a first beta version and use it with great care, particularly because of the stateless nature of the cluster. For now, I recommend to manually validate that you can get identical results using this cluster type with what you get from using the classical parallel::makeCluster() cluster type.

Usage

makeClusterFuture(specs = nbrOfWorkers(), ...)

Arguments

Value

Returns a parallel cluster object of class FutureCluster.

Future Clusters are Stateless

Traditionally, a cluster nodes has a one-to-one mapping to a cluster worker process. For example, cl <- makeCluster(2, type = "PSOCK") launches two parallel worker processes in the background, where cluster node cl[[1]] maps to worker #1 and node cl[[2]] to worker #2, and that never changes through the lifespan of these workers. This one-to-one mapping allows for deterministic configuration of workers. For examples, some code may assign globals with values specific to each worker, e.g. clusterEvalQ(cl[1], { a <- 3.14 }) and clusterEvalQ(cl[2], { a <- 2.71 }).

In contrast, there is no one-to-one mapping between cluster nodes and the parallel workers when using a future cluster. This is because we cannot make assumptions on where are parallel task will be processed. Where a parallel task is processes is up to the future backend to decide - some backends do this deterministically, whereas others other resolves task at the first available worker. Also, the worker processes might be transient for some future backends, i.e. the only exist for the life-span of the parallel task and then terminates.

Because of this, one must not rely in node-specific behaviors, because that concept does not make sense with a future cluster. To protect against this, any attempt to address a subset of future cluster nodes, results in an error, e.g. clusterEvalQ(cl[1], ...), clusterEvalQ(cl[1:2], ...), and clusterEvalQ(cl[2:1], ...) in the above example will all give an error.

Exceptions to the latter limitation are clusterSetRNGStream() and clusterExport(), which can be safely used with future clusters. See below for more details. If clusterEvalQ() is called, the call is ignored, and a warning is produced.

clusterSetRNGStream

parallel::clusterSetRNGStream() distributes "L'Ecuyer-CMRG" RNG streams to the cluster nodes, which record them such that the next round of futures will use them. When used, the RNG state after the futures are resolved are recorded accordingly, such that the next round again of future will use those, and so on. This strategy makes sure clusterSetRNGStream() has the expected effect although futures are stateless.

clusterExport

parallel::clusterExport() assign values to the cluster nodes. Specifically, these values are recorded and are used as globals for all futures created there on.

Examples

plan(multisession)
cl <- makeClusterFuture()

parallel::clusterSetRNGStream(cl)

y <- parallel::parLapply(cl, 11:13, function(x) {
  message("Process ID: ", Sys.getpid())
  mean(rnorm(n = x))
})
str(y)

plan(sequential)

Mandelbrot convergence counts

Description

Mandelbrot convergence counts

Usage

mandelbrot(...)

## S3 method for class 'matrix'
mandelbrot(Z, maxIter = 200L, tau = 2, ...)

## S3 method for class 'numeric'
mandelbrot(
  xmid = -0.75,
  ymid = 0,
  side = 3,
  resolution = 400L,
  maxIter = 200L,
  tau = 2,
  ...
)

Arguments

Value

Returns an integer matrix (of class Mandelbrot) with non-negative counts.

Author(s)

The internal Mandelbrot algorithm was inspired by and adopted from similar GPL code of Martin Maechler available from ftp://stat.ethz.ch/U/maechler/R/ on 2005-02-18 (sic!).

Examples

counts <- mandelbrot(xmid = -0.75, ymid = 0, side = 3)
str(counts)
## Not run: 
plot(counts)

## End(Not run)

## Not run: 
demo("mandelbrot", package = "future", ask = FALSE)

## End(Not run)

Create a multicore future whose value will be resolved asynchronously in a forked parallel process

Description

WARNING: This function must never be called. It may only be used with plan()

Usage

multicore(
  ...,
  workers = availableCores(constraints = "multicore"),
  gc = FALSE,
  earlySignal = FALSE,
  envir = parent.frame()
)

Arguments

Details

A multicore future is a future that uses multicore evaluation, which means that its value is computed and resolved in parallel in another process.

This function is must not be called directly. Instead, the typical usages are:

# Evaluate futures in parallel on the local machine via as many forked
# processes as available to the current R process
plan(multicore)

# Evaluate futures in parallel on the local machine via two forked processes
plan(multicore, workers = 2)

Value

A Future. If workers == 1, then all processing using done in the current/main R session and we therefore fall back to using a sequential future. To override this fallback, use workers = I(1). This is also the case whenever multicore processing is not supported, e.g. on Windows.

Support for forked ("multicore") processing

Not all operating systems support process forking and thereby not multicore futures. For instance, forking is not supported on Microsoft Windows. Moreover, process forking may break some R environments such as RStudio. Because of this, the future package disables process forking also in such cases. See parallelly::supportsMulticore() for details. Trying to create multicore futures on non-supported systems or when forking is disabled will result in multicore futures falling back to becoming sequential futures. If used in RStudio, there will be an informative warning:

> plan(multicore)
Warning message:
In supportsMulticoreAndRStudio(...) :
  [ONE-TIME WARNING] Forked processing ('multicore') is not supported when
running R from RStudio because it is considered unstable. For more details,
how to control forked processing or not, and how to silence this warning in
future R sessions, see ?parallelly::supportsMulticore

See Also

For processing in multiple background R sessions, see multisession futures.

For alternative future backends, see the 'A Future for R: Available Future Backends' vignette and https://www.futureverse.org/backends.html.

Use parallelly::availableCores() to see the total number of cores that are available for the current R session. Use availableCores("multicore") > 1L to check whether multicore futures are supported or not on the current system.

Examples

## Use multicore futures
plan(multicore)

## A global variable
a <- 0

## Create future (explicitly)
f <- future({
  b <- 3
  c <- 2
  a * b * c
})

## A multicore future is evaluated in a separate forked
## process.  Changing the value of a global variable
## will not affect the result of the future.
a <- 7
print(a)

v <- value(f)
print(v)
stopifnot(v == 0)

Create a multisession future whose value will be resolved asynchronously in a parallel R session

Description

WARNING: This function must never be called. It may only be used with plan()

Usage

multisession(
  ...,
  workers = availableCores(),
  lazy = FALSE,
  rscript_libs = .libPaths(),
  gc = FALSE,
  earlySignal = FALSE,
  envir = parent.frame()
)

Arguments

Details

A multisession future is a future that uses multisession evaluation, which means that its value is computed and resolved in parallel in another R session.

This function is must not be called directly. Instead, the typical usages are:

# Evaluate futures in parallel on the local machine via as many background
# processes as available to the current R process
plan(multisession)

# Evaluate futures in parallel on the local machine via two background
# processes
plan(multisession, workers = 2)

The background R sessions (the "workers") are created using makeClusterPSOCK().

For the total number of R sessions available including the current/main R process, see parallelly::availableCores().

A multisession future is a special type of cluster future.

Value

A MultisessionFuture. If workers == 1, then all processing is done in the current/main R session and we therefore fall back to using a lazy future. To override this fallback, use workers = I(1).

See Also

For processing in multiple forked R sessions, see multicore futures.

Use parallelly::availableCores() to see the total number of cores that are available for the current R session.

Examples

## Use multisession futures
plan(multisession)

## A global variable
a <- 0

## Create future (explicitly)
f <- future({
  b <- 3
  c <- 2
  a * b * c
})

## A multisession future is evaluated in a separate R session.
## Changing the value of a global variable will not affect
## the result of the future.
a <- 7
print(a)

v <- value(f)
print(v)
stopifnot(v == 0)

## Explicitly close multisession workers by switching plan
plan(sequential)

Get the number of workers available

Description

Get the number of workers available

Usage

nbrOfWorkers(evaluator = NULL)

nbrOfFreeWorkers(evaluator = NULL, background = FALSE, ...)

Arguments

Value

nbrOfWorkers() returns a positive number in {1, 2, 3, ...}, which for some future backends may also be +Inf.

nbrOfFreeWorkers() returns a non-negative number in {0, 1, 2, 3, ...} which is less than or equal to nbrOfWorkers().

Examples

plan(multisession)
nbrOfWorkers()  ## == availableCores()

plan(sequential)
nbrOfWorkers()  ## == 1

Creates a connection to the system null device

Description

Creates a connection to the system null device

Usage

nullcon()

Value

Returns a open, binary base::connection().

Functions Moved to 'parallelly'

Description

The following function used to be part of future but has since been migrated to parallelly. The migration started with future 1.20.0 (November 2020). They were moved because they are also useful outside of the future framework.

Details

If you are using any of these from the future package, please switch to use the ones from the parallelly package. Thank you!

• parallelly::as.cluster()

• parallelly::autoStopCluster() (no longer re-exported)

• parallelly::availableCores()

• parallelly::availableWorkers()

• parallelly::makeClusterMPI() (no longer re-exported)

• parallelly::makeClusterPSOCK()

• parallelly::makeNodePSOCK() (no longer re-exported)

• parallelly::supportsMulticore()

For backward-compatible reasons, some of these functions remain available as exact copies also from this package (as re-exports), e.g.

cl <- parallelly::makeClusterPSOCK(2)

can still be accessed as:

cl <- future::makeClusterPSOCK(2)

Note that it is the goal to remove all of the above from this package.

Writes and Reads 'immediateCondition' RDS Files

Description

Writes and Reads 'immediateCondition' RDS Files

Usage

readImmediateConditions(
  path = immediateConditionsPath(rootPath = rootPath),
  rootPath = tempdir(),
  pattern = "[.]rds$",
  include = getOption("future.relay.immediate", "immediateCondition"),
  signal = FALSE,
  remove = TRUE
)

saveImmediateCondition(
  cond,
  path = immediateConditionsPath(rootPath = rootPath),
  rootPath = tempdir()
)

Arguments

Value

readImmediateConditions() returns an unnamed base::list of named lists with elements condition and signaled, where the condition elements hold immediateCondition objects.

saveImmediateCondition() returns, invisibly, the pathname of the RDS written.

Request a core for multicore processing

Description

If no cores are available, the current process blocks until a core is available.

Usage

requestCore(
  await,
  workers = availableCores(constraints = "multicore"),
  timeout,
  delta,
  alpha
)

Arguments

Value

Invisible TRUE. If no cores are available after extensive waiting, then a timeout error is thrown.

Reset a finished, failed, canceled, or interrupted future to a lazy future

Description

A future that has successfully completed, canceled, interrupted, or has failed due to an error, can be relaunched after resetting it.

Usage

reset(x, ...)

Arguments

Details

A lazy, vanilla Future can be reused in another R session. For instance, if we do:

library(future)
a <- 2
f <- future(42 * a, lazy = TRUE)
saveRDS(f, "myfuture.rds")

Then we can read and evaluate the future in another R session using:

library(future)
f <- readRDS("myfuture.rds")
v <- value(f)
print(v)
#> [1] 84

Value

reset() returns a lazy, vanilla Future that can be relaunched. Resetting a running future results in a FutureError.

Examples

## Like mean(), but fails 90% of the time
shaky_mean <- function(x) {
  if (as.double(Sys.time()) %% 1 < 0.90) stop("boom")
  mean(x)
}

x <- rnorm(100)

## Calculate the mean of 'x' with a risk of failing randomly
f <- future({ shaky_mean(x) })

## Relaunch until success
repeat({
  v <- tryCatch(value(f), error = identity)
  if (!inherits(v, "error")) break
  message("Resetting failed future, and retry in 0.1 seconds")
  f <- reset(f)
  Sys.sleep(0.1)
})
cat("mean:", v, "\n")

Free up active background workers

Description

Free up active background workers

Usage

resetWorkers(x, ...)

Arguments

Details

This function will resolve any active futures that is currently being evaluated on background workers.

Examples

resetWorkers(plan())

Resolve one or more futures synchronously

Description

This function provides an efficient mechanism for waiting for multiple futures in a container (e.g. list or environment) to be resolved while in the meanwhile retrieving values of already resolved futures.

Usage

resolve(
  x,
  idxs = NULL,
  recursive = 0,
  result = FALSE,
  stdout = FALSE,
  signal = FALSE,
  force = FALSE,
  sleep = getOption("future.wait.interval", 0.01),
  ...
)

Arguments

Details

This function is resolves synchronously, i.e. it blocks until x and any containing futures are resolved.

Value

Returns x (regardless of subsetting or not). If signal is TRUE and one of the futures produces an error, then that error is produced.

See Also

To resolve a future variable, first retrieve its Future object using futureOf(), e.g. resolve(futureOf(x)).

Check whether a future is resolved or not

Description

Check whether a future is resolved or not

Usage

resolved(x, ...)

Arguments

Details

This method needs to be implemented by the class that implement the Future API. The implementation should return either TRUE or FALSE and must never throw an error (except for FutureError:s which indicate significant, often unrecoverable infrastructure problems). It should also be possible to use the method for polling the future until it is resolved (without having to wait infinitely long), e.g. while (!resolved(future)) Sys.sleep(5).

Value

A logical of the same length and dimensions as x. Each element is TRUE unless the corresponding element is a non-resolved future in case it is FALSE.

Get the results of a resolved future

Description

Get the results of a resolved future

Usage

## S3 method for class 'Future'
result(future, ...)

Arguments

Details

This function is only part of the backend Future API. This function is not part of the frontend Future API.

Value

The FutureResult object.

Run a future

Description

Run a future

Usage

## S3 method for class 'Future'
run(future, ...)

Arguments

Details

This function can only be called once per future. Further calls will result in an informative error. If a future is not run when its value is queried, then it is run at that point.

Value

The Future object.

Robustly Saves an Object to RDS File Atomically

Description

Robustly Saves an Object to RDS File Atomically

Usage

save_rds(object, pathname, ...)

Arguments

Details

Uses base::saveRDS internally but writes the object atomically by first writing to a temporary file which is then renamed.

Value

The pathname of the RDS written.

Create a sequential future whose value will be in the current R session

Description

WARNING: This function must never be called. It may only be used with plan()

Usage

sequential(..., gc = FALSE, earlySignal = FALSE, envir = parent.frame())

Arguments

Details

A sequential future is a future that is evaluated sequentially in the current R session similarly to how R expressions are evaluated in R. The only difference to R itself is that globals are validated by default just as for all other types of futures in this package.

This function is must not be called directly. Instead, the typical usages are:

# Evaluate futures sequentially in the current R process
plan(sequential)

Value

A Future.

Examples

## Use sequential futures
plan(sequential)

## A global variable
a <- 0

## Create a sequential future
f <- future({
  b <- 3
  c <- 2
  a * b * c
})

## Since 'a' is a global variable in future 'f' which
## is eagerly resolved (default), this global has already
## been resolved / incorporated, and any changes to 'a'
## at this point will _not_ affect the value of 'f'.
a <- 7
print(a)

v <- value(f)
print(v)
stopifnot(v == 0)

Outputs details on the current R session

Description

Outputs details on the current R session

Usage

sessionDetails(env = FALSE)

Arguments

Value

Invisibly a list of all details.

Signals Captured Conditions

Description

Captured conditions that meet the include and exclude requirements are signaled in the order as they were captured.

Usage

signalConditions(
  future,
  include = "condition",
  exclude = NULL,
  resignal = TRUE,
  ...
)

Arguments

Value

Returns the Future where conditioned that were signaled have been flagged to have been signaled.

See Also

Conditions are signaled by signalCondition().

Place a sticky-globals environment immediately after the global environment

Description

Place a sticky-globals environment immediately after the global environment

Usage

sticky_globals(erase = FALSE, name = "future:sticky_globals", pos = 2L)

Arguments

Value

(invisible; environment) The environment.

Get number of cores currently used

Description

Get number of children (and don't count the current process) used by the current R session. The number of children is the total number of subprocesses launched by this process that are still running and whose values have yet not been collected.

Usage

usedCores()

Value

A non-negative integer.

The value of a future or the values of all elements in a container

Description

Gets the value of a future or the values of all elements (including futures) in a container such as a list, an environment, or a list environment. If one or more futures is unresolved, then this function blocks until all queried futures are resolved.

Usage

value(...)

## S3 method for class 'Future'
value(future, stdout = TRUE, signal = TRUE, drop = FALSE, ...)

## S3 method for class 'list'
value(
  x,
  idxs = NULL,
  recursive = 0,
  reduce = NULL,
  stdout = TRUE,
  signal = TRUE,
  cancel = TRUE,
  interrupt = cancel,
  inorder = TRUE,
  drop = FALSE,
  force = TRUE,
  sleep = getOption("future.wait.interval", 0.01),
  ...
)

## S3 method for class 'listenv'
value(
  x,
  idxs = NULL,
  recursive = 0,
  reduce = NULL,
  stdout = TRUE,
  signal = TRUE,
  cancel = TRUE,
  interrupt = cancel,
  inorder = TRUE,
  drop = FALSE,
  force = TRUE,
  sleep = getOption("future.wait.interval", 0.01),
  ...
)

## S3 method for class 'environment'
value(x, ...)

Arguments

Value

value() of a Future object returns the value of the future, which can be any type of R object.

value() of a list, an environment, or a list environment returns an object with the same number of elements and of the same class. Names and dimension attributes are preserved, if available. All future elements are replaced by their corresponding value() values. For all other elements, the existing object is kept as-is.

If signal is TRUE and one of the futures produces an error, then that error is relayed. Any remaining, non-resolved futures in x are canceled, prior to signaling such an error.

Examples

## ------------------------------------------------------
## A single future
## ------------------------------------------------------
x <- sample(100, size = 50)
f <- future(mean(x))
v <- value(f)
message("The average of 50 random numbers in [1,100] is: ", v)



## ------------------------------------------------------
## Ten futures
## ------------------------------------------------------
xs <- replicate(10, { list(sample(100, size = 50)) })
fs <- lapply(xs, function(x) { future(mean(x)) })

## The 10 values as a list (because 'fs' is a list)
vs <- value(fs)
message("The ten averages are:")
str(vs)

## The 10 values as a vector (by manually unlisting)
vs <- value(fs)
vs <- unlist(vs)
message("The ten averages are: ", paste(vs, collapse = ", "))

## The values as a vector (by reducing)
vs <- value(fs, reduce = `c`)
message("The ten averages are: ", paste(vs, collapse = ", "))

## Calculate the sum of the averages (by reducing)
total <- value(fs, reduce = `sum`)
message("The sum of the ten averages is: ", total)

Evaluate an expression using a temporarily set future plan

Description

This function allows the user to plan the future, more specifically, it specifies how future():s are resolved, e.g. sequentially or in parallel.

Usage

## S3 method for class 'FutureStrategyList'
with(data, expr, ..., local = FALSE, envir = parent.frame(), .cleanup = NA)

plan(
  strategy = NULL,
  ...,
  substitute = TRUE,
  .skip = FALSE,
  .call = TRUE,
  .cleanup = NA,
  .init = TRUE
)

tweak(strategy, ..., penvir = parent.frame())

Arguments

Details

The default backend is sequential, but another one can be set using plan(), e.g. plan(multisession) will launch parallel workers running in the background, which then will be used to resolve future. To shut down background workers launched this way, call plan(sequential).

Value

The value of the expression evaluated (invisibly).

plan() returns a the previous plan invisibly if a new future backend is chosen, otherwise it returns the current one visibly.

a future function.

Built-in evaluation strategies

The future package provides the following built-in backends:

sequential:

Resolves futures sequentially in the current R process, e.g. plan(sequential).

multisession:

Resolves futures asynchronously (in parallel) in separate R sessions running in the background on the same machine, e.g. plan(multisession) and plan(multisession, workers = 2).

multicore:

Resolves futures asynchronously (in parallel) in separate forked R processes running in the background on the same machine, e.g. plan(multicore) and plan(multicore, workers = 2). This backend is not supported on Windows.

cluster:

Resolves futures asynchronously (in parallel) in separate R sessions running typically on one or more machines, e.g. plan(cluster), plan(cluster, workers = 2), and plan(cluster, workers = c("n1", "n1", "n2", "server.remote.org")).

Other evaluation strategies available

In addition to the built-in ones, additional parallel backends are implemented in future-backend packages future.callr and future.mirai that leverage R package callr and mirai:

callr:

Similar to multisession, this resolved futures in parallel in background R sessions on the local machine via the callr package, e.g. plan(future.callr::callr) and plan(future.callr::callr, workers = 2). The difference is that each future is processed in a fresh parallel R worker, which is automatically shut down as soon as the future is resolved. This can help decrease the overall memory. Moreover, contrary to multisession, callr does not rely on socket connections, which means it is not limited by the number of connections that R can have open at any time.

mirai_multisession:

Similar to multisession, this resolved futures in parallel in background R sessions on the local machine via the mirai package, e.g. plan(future.mirai::mirai_multisession) and plan(future.mirai::mirai_multisession, workers = 2).

mirai_cluster:

Similar to cluster, this resolved futures in parallel via pre-configured R mirai daemon processes, e.g. plan(future.mirai::mirai_cluster).

Another example is the future.batchtools package, which leverages batchtools package, to resolve futures via high-performance compute (HPC) job schedulers, e.g. LSF, Slurm, TORQUE/PBS, Grid Engine, and OpenLava;

batchtools_slurm:

The backend resolved futures via the Slurm scheduler, e.g. plan(future.batchtools::batchtools_slurm).

batchtools_torque:

The backend resolved futures via the TORQUE/PBS scheduler, e.g. plan(future.batchtools::batchtools_torque).

batchtools_sge:

The backend resolved futures via the Grid Engine (SGE, AGE) scheduler, e.g. plan(future.batchtools::batchtools_sge).

batchtools_lsf:

The backend resolved futures via the Load Sharing Facility (LSF) scheduler, e.g. plan(future.batchtools::batchtools_lsf).

batchtools_openlava:

The backend resolved futures via the OpenLava scheduler, e.g. plan(future.batchtools::batchtools_openlava).

For package developers

Please refrain from modifying the future backend inside your packages / functions, i.e. do not call plan() in your code. Instead, leave the control on what backend to use to the end user. This idea is part of the core philosophy of the future framework—as a developer you can never know what future backends the user have access to. Moreover, by not making any assumptions about what backends are available, your code will also work automatically with any new backends developed after you wrote your code.

If you think it is necessary to modify the future backend within a function, then make sure to undo the changes when exiting the function. This can be achieved by using with(plan(...), local = TRUE), e.g.

  my_fcn <- function(x) {
    with(plan(multisession), local = TRUE)
    y <- analyze(x)
    summarize(y)
  }

This is important because the end-user might have already set the future strategy elsewhere for other purposes and will most likely not known that calling your function will break their setup. Remember, your package and its functions might be used in a greater context where multiple packages and functions are involved and those might also rely on the future framework, so it is important to avoid stepping on others' toes.

Using plan() in scripts and vignettes

When writing scripts or vignettes that use futures, try to place any call to plan() as far up (i.e. as early on) in the code as possible. This will help users to quickly identify where the future plan is set up and allow them to modify it to their computational resources. Even better is to leave it to the user to set the plan() prior to source():ing the script or running the vignette. If a ‘.future.R’ exists in the current directory and / or in the user's home directory, it is sourced when the future package is loaded. Because of this, the ‘.future.R’ file provides a convenient place for users to set the plan(). This behavior can be controlled via an R option—see future options for more details.

See Also

Use plan() to set a future to become the new default strategy.

Examples

# Evaluate a future using the 'multisession' plan
with(plan(multisession, workers = 2), {
  f <- future(Sys.getpid())
  w_pid <- value(f)
})
print(c(main = Sys.getpid(), worker = w_pid))



# Evaluate a future locally using the 'multisession' plan
local({
  with(plan(multisession, workers = 2), local = TRUE)

  f <- future(Sys.getpid())
  w_pid <- value(f)
  print(c(main = Sys.getpid(), worker = w_pid))
})


a <- b <- c <- NA_real_

# An sequential future
plan(sequential)
f <- future({
  a <- 7
  b <- 3
  c <- 2
  a * b * c
})
y <- value(f)
print(y)
str(list(a = a, b = b, c = c)) ## All NAs


# A sequential future with lazy evaluation
plan(sequential)
f <- future({
  a <- 7
  b <- 3
  c <- 2
  a * b * c
}, lazy = TRUE)
y <- value(f)
print(y)
str(list(a = a, b = b, c = c)) ## All NAs


# A multicore future (specified as a string)
plan("multicore")
f <- future({
  a <- 7
  b <- 3
  c <- 2
  a * b * c
})
y <- value(f)
print(y)
str(list(a = a, b = b, c = c)) ## All NAs

## Multisession futures gives an error on R CMD check on
## Windows (but not Linux or macOS) for unknown reasons.
## The same code works in package tests.


# A multisession future (specified via a string variable)
plan("future::multisession")
f <- future({
  a <- 7
  b <- 3
  c <- 2
  a * b * c
})
y <- value(f)
print(y)
str(list(a = a, b = b, c = c)) ## All NAs




## Explicitly specifying number of workers
## (default is parallelly::availableCores())
plan(multicore, workers = 2)
message("Number of parallel workers: ", nbrOfWorkers())


## Explicitly close multisession workers by switching plan
plan(sequential)

Options used for futures

Description

Below are the R options and environment variables that are used by the future 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.

Packages must not change future options

Just like for other R options, as a package developer you must not change any of the below ⁠future.*⁠ options. Only the end-user should set these. If you find yourself having to tweak one of the options, make sure to undo your changes immediately afterward. For example, if you want to bump up the future.globals.maxSize limit when creating a future, use something like the following inside your function:

oopts <- options(future.globals.maxSize = 1.0 * 1e9)  ## 1.0 GB
on.exit(options(oopts))
f <- future({ expr })  ## Launch a future with large objects

Options for controlling futures

future.plan:

(character string or future function) Default future backend used unless otherwise specified via plan(). This will also be the future plan set when calling plan("default"). If not specified, this option may be set when the future package is loaded if command-line option --parallel=ncores (short ⁠-p ncores⁠) is specified; if ncores > 1, then option future.plan is set to multisession otherwise sequential (in addition to option mc.cores being set to ncores, if ncores >= 1). (Default: sequential)

future.globals.maxSize:

(numeric) Maximum allowed total size (in bytes) of global variables identified. This is used to protect against exporting too large objects to parallel workers by mistake. Transferring large objects over a network, or over the internet, can be slow and therefore introduce a large bottleneck that increases the overall processing time. It can also result in large egress or ingress costs, which may exist on some systems. If set of +Inf, then the check for large globals is skipped. (Default: 500 * 1024 ^ 2 = 500 MiB)

future.globals.onReference: (beta feature - may change)

(character string) Controls whether the identified globals should be scanned for so called references (e.g. external pointers and connections) or not. It is unlikely that another R process ("worker") can use a global that uses a internal reference of the master R process—we call such objects non-exportable globals. If this option is "error", an informative error message is produced if a non-exportable global is detected. If "warning", a warning is produced, but the processing will continue; it is likely that the future will be resolved with a run-time error unless processed in the master R process (e.g. plan(sequential) and plan(multicore)). If "ignore", no scan is performed. (Default: "ignore" but may change)

future.resolve.recursive:

(integer) An integer specifying the maximum recursive depth to which futures should be resolved. If negative, nothing is resolved. If 0, only the future itself is resolved. If 1, the future and any of its elements that are futures are resolved, and so on. If +Inf, infinite search depth is used. (Default: 0)

future.onFutureCondition.keepFuture:

(logical) If TRUE, a FutureCondition keeps a copy of the Future object that triggered the condition. If FALSE, it is dropped. (Default: TRUE)

future.wait.timeout:

(numeric) Maximum waiting time (in seconds) for a future to resolve or for a free worker to become available before a timeout error is generated. (Default: 30 * 24 * 60 * 60 (= 30 days))

future.wait.interval:

(numeric) Initial interval (in seconds) between polls. This controls the polling frequency for finding an available worker when all workers are currently busy. It also controls the polling frequency of resolve(). (Default: 0.01 = 1 ms)

future.wait.alpha:

(numeric) Positive scale factor used to increase the interval after each poll. (Default: 1.01)

Options for built-in sanity checks

Ideally, the evaluation of a future should have no side effects. To protect against unexpected side effects, the future framework comes with a set of built-in tools for checking against this. Below R options control these built-in checks and what should happen if they fail. You may modify them for troubleshooting purposes, but please refrain from disabling these checks when there is an underlying problem that should be fixed.

Beta features: Please consider these checks to be "under construction".

future.connections.onMisuse:

(character string) A future must close any connections it opens and must not close connections it did not open itself. If such misuse is detected and this option is set to "error", then an informative error is produced. If it is set to "warning", a warning is produced. If"ignore", no check is performed. (Default: "warning")

future.defaultDevice.onMisuse:

(character string) A future must open graphics devices explicitly, if it creates new plots. It should not rely on the default graphics device that is given by R option "default", because that rarely does what is intended. If such misuse is detected and this option is set to "error", then an informative error is produced. If it is set to "warning", a warning is produced. If"ignore", no check is performed. (Default: "warning")

future.devices.onMisuse:

(character string) A future must close any graphics devices it opens and must not close devices it did not open itself. If such misuse is detected and this option is set to "error", then an informative error is produced. If it is set to "warning", a warning is produced. If"ignore", no check is performed. (Default: "warning")

future.globalenv.onMisuse:

(character string) Assigning variables to the global environment for the purpose of using the variable at a later time makes no sense with futures, because the next the future may be evaluated in different R process. To protect against mistakes, the future framework attempts to detect when variables are added to the global environment. If this is detected, and this option is set to "error", then an informative error is produced. If "warning", then a warning is produced. If "ignore", no check is performed. (Default: "ignore")

future.rng.onMisuse:

(character string) If random numbers are used in futures, then parallel RNG should be declared in order to get statistical sound RNGs. You can declare this by specifying future argument seed = TRUE. The defaults in the future framework assume that no random number generation (RNG) is taken place in the future expression because L'Ecuyer-CMRG RNGs come with an unnecessary overhead if not needed. To protect against mistakes of not declaring use of the RNG, the future framework detects when random numbers were used despite not declaring such use. If this is detected, and this options is set "error", then an informative error is produced. If "warning", then a warning is produced. If "ignore", no check is performed. (Default: "warning")

Options for debugging futures

future.debug:

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

Options for controlling package startup

future.startup.script:

(character vector or a logical) Specifies zero of more future startup scripts to be sourced when the future package is attached. It is only the first existing script that is sourced. If none of the specified files exist, nothing is sourced—there will be neither a warning nor an error. If this option is not specified, environment variable R_FUTURE_STARTUP_SCRIPT is considered, where multiple scripts may be separated by either a colon (:) or a semicolon (⁠;⁠). If neither is set, or either is set to TRUE, the default is to look for a ‘.future.R’ script in the current directory and then in the user's home directory. To disable future startup scripts, set the option or the environment variable to FALSE. Importantly, this option is always set to FALSE if the future package is loaded as part of a future expression being evaluated, e.g. in a background process. In order words, they are sourced in the main R process but not in future processes. (Default: TRUE in main R process and FALSE in future processes / during future evaluation)

future.cmdargs:

(character vector) Overrides commandArgs() when the future package is loaded.

Options for configuring low-level system behaviors

future.fork.multithreading.enable (beta feature - may change):

(logical) Enable or disable multi-threading while using forked parallel processing. If FALSE, different multi-thread library settings are overridden such that they run in single-thread mode. Specifically, multi-threading will be disabled for OpenMP (which requires the RhpcBLASctl package) and for RcppParallel. If TRUE, or not set (the default), multi-threading is allowed. Parallelization via multi-threaded processing (done in native code by some packages and external libraries) while at the same time using forked (aka "multicore") parallel processing is known to unstable. Note that this is not only true when using plan(multicore) but also when using, for instance, mclapply() of the parallel package. (Default: not set)

future.output.windows.reencode:

(logical) Enable or disable re-encoding of UTF-8 symbols that were incorrectly encoded while captured. In R (< 4.2.0) and on older versions of MS Windows, R cannot capture UTF-8 symbols as-is when they are captured from the standard output. For examples, a UTF-8 check mark symbol ("\u2713") would be relayed as "<U+2713>" (a string with eight ASCII characters). Setting this option to TRUE will cause value() to attempt to recover the intended UTF-8 symbols from ⁠<U+nnnn>⁠ string components, if, and only if, the string was captured by a future resolved on MS Windows. (Default: TRUE)

Options for demos

future.demo.mandelbrot.region:

(integer) Either a named list of mandelbrot() arguments or an integer in {1, 2, 3} specifying a predefined Mandelbrot region. (Default: 1L)

future.demo.mandelbrot.nrow:

(integer) Number of rows and columns of tiles. (Default: 3L)

Deprecated or for internal prototyping

The following options exists only for troubleshooting purposes and must not be used in production. If used, there is a risk that the results are non-reproducible if processed elsewhere. To lower the risk of them being used by mistake, they are marked as deprecated and will produce warnings if set.

future.globals.onMissing:

(character string) Action to take when non-existing global variables ("globals" or "unknowns") are identified when the future is created. If "error", an error is generated immediately. If "ignore", no action is taken and an attempt to evaluate the future expression will be made. The latter is useful when there is a risk for false-positive globals being identified, e.g. when future expression contains non-standard evaluation (NSE). (Default: "ignore")

future.globals.method:

(character string) Method used to identify globals. For details, see globalsOf(). (Default: "ordered")

future.globals.resolve:

(logical) If TRUE, globals that are Future objects (typically created as explicit futures) will be resolved and have their values (using value()) collected. Because searching for unresolved futures among globals (including their content) can be expensive, the default is not to do it and instead leave it to the run-time checks that assert proper ownership when resolving futures and collecting their values. (Default: FALSE)

Environment variables that set R options

All of the above R future.* options can be set by corresponding environment variable R_FUTURE_* when the future package is loaded. This means that those environment variables must be set before the future package is loaded in order to have an effect. For example, if R_FUTURE_RNG_ONMISUSE="ignore", then option future.rng.onMisuse is set to "ignore" (character string). Similarly, if R_FUTURE_GLOBALS_MAXSIZE="50000000", then option future.globals.maxSize is set to 50000000 (numeric).

Options moved to the 'parallelly' package

Several functions have been moved to the parallelly package:

• parallelly::availableCores()

• parallelly::availableWorkers()

• parallelly::makeClusterMPI()

• parallelly::makeClusterPSOCK()

• parallelly::makeNodePSOCK()

• parallelly::supportsMulticore()

The options and environment variables controlling those have been adjusted accordingly to have different prefixes. For example, option future.fork.enable has been renamed to parallelly.fork.enable and the corresponding environment variable R_FUTURE_FORK_ENABLE has been renamed to R_PARALLELLY_FORK_ENABLE. For backward compatibility reasons, the parallelly package will support both versions for a long foreseeable time. See the parallelly::parallelly.options page for the settings.

See Also

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

Examples

# Allow at most 5 MB globals per futures
options(future.globals.maxSize = 5e6)

# Be strict; catch all RNG mistakes
options(future.rng.onMisuse = "error")