Minimalist Async Evaluation Framework for R • mirai

Minimalist async evaluation framework for R.

Lightweight parallel code execution and distributed computing.

Designed for simplicity, a ‘mirai’ evaluates an R expression asynchronously, on local or network resources, resolving automatically upon completion.

Features efficient task scheduling, scalability beyond R connection limits, and transports faster than TCP/IP for inter-process communications, courtesy of ‘nanonext’ and ‘NNG’ (Nanomsg Next Gen).

mirai() returns a ‘mirai’ object immediately. ‘mirai’ (未来みらい) is Japanese for ‘future’.

mirai has a tiny pure R code base, relying solely on nanonext, a high-performance binding for the ‘NNG’ (Nanomsg Next Gen) C library with zero package dependencies.

Installation
Example 1: Compute-intensive Operations
Example 2: I/O-bound Operations
Example 3: Resilient Pipelines
Daemons: Local Persistent Processes
Distributed Computing: Remote Servers
Compute Profiles
Errors, Interrupts and Timeouts
Deferred Evaluation Pipe
Integrations with Crew, Targets, Shiny
Thanks
Links

Installation

Install the latest release from CRAN:

install.packages("mirai")

or the development version from rOpenSci R-universe:

install.packages("mirai", repos = "https://shikokuchuo.r-universe.dev")

« Back to ToC

Example 1: Compute-intensive Operations

Use case: minimise execution times by performing long-running tasks concurrently in separate processes.

Multiple long computes (model fits etc.) can be performed in parallel on available computing cores.

Use mirai() to evaluate an expression asynchronously in a separate, clean R process.

A ‘mirai’ object is returned immediately.

library(mirai)

m <- mirai(
  {
    res <- rnorm(n) + m
    res / rev(res)
  },
  m = runif(1),
  n = 1e8
)

m
#> < mirai >
#>  - $data for evaluated result

Above, all specified name = value pairs are passed through to the ‘mirai’.

The ‘mirai’ yields an ‘unresolved’ logical NA value whilst the async operation is ongoing.

m$data
#> 'unresolved' logi NA

Upon completion, the ‘mirai’ resolves automatically to the evaluated result.

m$data |> str()
#>  num [1:100000000] 2.85 3.63 0.25 1.23 -1.45 ...

Alternatively, explicitly call and wait for the result using call_mirai().

call_mirai(m)$data |> str()
#>  num [1:100000000] 2.85 3.63 0.25 1.23 -1.45 ...

For easy programmatic use of mirai(), ‘.expr’ accepts a pre-constructed language object, and also a list of named arguments passed via ‘.args’. So, the following would be equivalent to the above:

expr <- quote({
  res <- rnorm(n) + m
  res / rev(res)
})

args <- list(m = runif(1), n = 1e8)

m <- mirai(.expr = expr, .args = args)

call_mirai(m)$data |> str()
#>  num [1:100000000] -2.1689 3.1048 0.1179 -1.269 0.0924 ...

« Back to ToC

Example 2: I/O-bound Operations

Use case: ensure execution flow of the main process is not blocked.

High-frequency real-time data cannot be written to file/database synchronously without disrupting the execution flow.

Cache data in memory and use mirai() to perform periodic write operations concurrently in a separate process.

Below, ‘.args’ is used to pass a list of objects already present in the calling environment to the mirai by name. This is an alternative use of ‘.args’, and may be combined with ... to also pass in name = value pairs.

library(mirai)

x <- rnorm(1e6)
file <- tempfile()

m <- mirai(write.csv(x, file = file), .args = list(x, file))

A ‘mirai’ object is returned immediately.

unresolved() may be used in control flow statements to perform actions which depend on resolution of the ‘mirai’, both before and after.

This means there is no need to actually wait (block) for a ‘mirai’ to resolve, as the example below demonstrates.

# unresolved() queries for resolution itself so no need to use it again within the while loop

while (unresolved(m)) {
  cat("while unresolved\n")
  Sys.sleep(0.5)
}
#> while unresolved
#> while unresolved

cat("Write complete:", is.null(m$data))
#> Write complete: TRUE

Now actions which depend on the resolution may be processed, for example the next write.

« Back to ToC

Example 3: Resilient Pipelines

Use case: isolating code that can potentially fail in a separate process to ensure continued uptime.

As part of a data science / machine learning pipeline, iterations of model training may periodically fail for stochastic and uncontrollable reasons (e.g. buggy memory management on graphics cards).

Running each iteration in a ‘mirai’ isolates this potentially-problematic code such that even if it does fail, it does not bring down the entire pipeline.

library(mirai)

run_iteration <- function(i) {
  
  if (runif(1) < 0.1) stop("random error", call. = FALSE) # simulates a stochastic error rate
  sprintf("iteration %d successful", i)
  
}

for (i in 1:10) {
  
  m <- mirai(run_iteration(i), .args = list(run_iteration, i))
  while (is_error_value(call_mirai(m)$data)) {
    cat(m$data, "\n")
    m <- mirai(run_iteration(i), .args = list(run_iteration, i))
  }
  cat(m$data, "\n")
  
}
#> iteration 1 successful 
#> iteration 2 successful 
#> iteration 3 successful 
#> Error: random error 
#> iteration 4 successful 
#> iteration 5 successful 
#> iteration 6 successful 
#> iteration 7 successful 
#> iteration 8 successful 
#> iteration 9 successful 
#> iteration 10 successful

Further, by testing the return value of each ‘mirai’ for errors, error-handling code is then able to automate recovery and re-attempts, as in the above example. Further details on error handling can be found in the section below.

The end result is a resilient and fault-tolerant pipeline that minimises downtime by eliminating interruptions of long computes.

« Back to ToC

Daemons: Local Persistent Processes

Daemons, or persistent background processes, may be set to receive ‘mirai’ requests.

This is potentially more efficient as new processes no longer need to be created on an ad hoc basis.

With Dispatcher (default)

Call daemons() specifying the number of daemons to launch.

daemons(6)
#> [1] 6

To view the current status, call daemons() with no arguments. This provides the number of active connections along with a matrix of statistics for each daemon.

daemons()
#> $connections
#> [1] 1
#> 
#> $daemons
#>                                                     online instance assigned complete
#> abstract://fa3b45a0e2d35f5f97b82898742aa0cbd198bec1      1        1        0        0
#> abstract://334ead7a516f80c4c69b45ecae45e99d1ecdf343      1        1        0        0
#> abstract://0620a1e9c722af1e357e710228aa20f9f7e41a1b      1        1        0        0
#> abstract://154b6c85497a8a38d5872e3c6b3df1f2a5771c77      1        1        0        0
#> abstract://94ed649b5329764d0f68946ade8b99650ca5b0d4      1        1        0        0
#> abstract://8cbba13ba2c6db84a51aade8c6f52ef3eeb272a1      1        1        0        0

The default dispatcher = TRUE creates a dispatcher() background process that connects to individual daemon processes on the local machine on behalf of the client. This ensures that tasks are dispatched efficiently on a first-in first-out (FIFO) basis to servers for processing. Tasks are queued at the dispatcher and sent to a daemon as soon as it can accept the task for immediate execution.

Dispatcher uses synchronisation primitives from nanonext, waiting upon rather than polling for tasks, which is efficient both in terms of consuming no resources while waiting, and also being fully synchronised with events (having no latency).

daemons(0)
#> [1] 0

Set the number of daemons to zero to reset. This reverts to the default of creating a new background process for each ‘mirai’ request.

Without Dispatcher

Alternatively, specifying dispatcher = FALSE, the background daemon processes connect directly to the client.

daemons(6, dispatcher = FALSE)
#> [1] 6

Requesting the status now shows 6 connections and 6 daemons.

daemons()
#> $connections
#> [1] 6
#> 
#> $daemons
#> [1] 6

This implementation sends tasks immediately, and ensures that tasks are evenly-distributed amongst daemons. This means that optimal scheduling is not guaranteed as the duration of tasks cannot be known a priori. As an example, tasks could be queued at a daemon behind a long-running task, whilst other daemons remain idle.

The advantage of this approach is that it is low-level and does not require an additional dispatcher process. It is well-suited to working with similar-length tasks, or where the number of concurrent tasks typically does not exceed available daemons.

daemons(0)
#> [1] 0

Set the number of daemons to zero to reset.

« Back to ToC

Distributed Computing: Remote Servers

The daemons interface may also be used to send tasks for computation to server processes on the network.

Call daemons() specifying ‘url’ as a character string the client network address and a port that is able to accept incoming connections.

The examples below use an illustrative local network IP address of ‘10.111.5.13’.

A port on the client also needs to be open and available for inbound connections from the local network, illustratively ‘5555’ in the examples below.

Connecting to Remote Servers Through Dispatcher

The default dispatcher = TRUE creates a background dispatcher() process on the local client machine, which listens to a vector of URLs that remote server() processes dial in to, with each server having its unique URL.

It is recommended to use a websocket URL starting ws:// instead of TCP in this scenario (used interchangeably with tcp://). A websocket URL supports a path after the port number, which can be made unique for each server. In this way a dispatcher can connect to an arbitrary number of servers over a single port.

# daemons(n = 4, url = "ws://10.111.5.13:5555")

daemons(n = 4, url = "ws://:5555")
#> [1] 4

Above, a single URL was supplied, along with n = 4 to specify that the dispatcher should listen at 4 URLs. In such a case, an integer sequence is automatically appended to the path /1 through /4 to produce these URLs.

Alternatively, supplying a vector of URLs allows the use of arbitrary port numbers / paths, e.g.:

# daemons(url = c("ws://:5555/cpu", "ws://:5555/gpu", "ws://:12560", "ws://:12560/2"))

Above, ‘n’ is not specified, in which case its value is inferred from the length of the ‘url’ vector supplied.

–

On the remote resource, server() may be called from an R session, or directly from a shell using Rscript. Each server instance should dial into one of the unique URLs that the dispatcher is listening to:

Rscript -e 'mirai::server("ws://10.111.5.13:5555/1")'
Rscript -e 'mirai::server("ws://10.111.5.13:5555/2")'
Rscript -e 'mirai::server("ws://10.111.5.13:5555/3")'
Rscript -e 'mirai::server("ws://10.111.5.13:5555/4")'

Note that daemons() should be set up on the client before launching server() on remote resources, otherwise the server instances will exit if a connection is not immediately available. Alternatievly specifying server(asyncdial = TRUE) will allow servers to wait (indefinitely) for a connection to become available.

–

Requesting status, on the client:

daemons()
#> $connections
#> [1] 1
#> 
#> $daemons
#>              online instance assigned complete
#> ws://:5555/1      1        1        0        0
#> ws://:5555/2      1        1        0        0
#> ws://:5555/3      1        1        0        0
#> ws://:5555/4      1        1        0        0

As per the local case, $connections will show the single connection to the dispatcher process, however $daemons will provide the matrix of statistics for the remote servers.

online shows as 1 when there is an active connection, or else 0 if a server has yet to connect or has disconnected.

instance increments by 1 every time there is a new connection at a URL. When this happens, the ‘assigned’ and ‘complete’ statistics reset to zero. This counter is designed to track new server instances connecting after previous ones have ended (due to time-outs etc.). ‘instance’ itself resets to zero if the URL is regenerated by saisei().

assigned shows the cumulative number of tasks assigned to the server instance by the dispatcher.

complete shows the cumulative number of tasks completed by the server instance.

The dispatcher automatically adjusts to the number of servers actually connected. Hence it is possible to dynamically scale up or down the number of servers according to requirements (limited to the ‘n’ URLs assigned at the dispatcher).

To reset all connections and revert to default behaviour:

daemons(0)
#> [1] 0

Closing the connection causes the dispatcher to exit automatically, and in turn all connected servers when their respective connections with the dispatcher are terminated.

Connecting to Remote Servers Directly

By specifying dispatcher = FALSE, remote servers connect directly to the client. The client listens at a single URL address, and distributes tasks to all connected server processes.

daemons(url = "tcp://10.111.5.13:0", dispatcher = FALSE)

Alternatively, simply supply a colon followed by the port number to listen on all interfaces on the local host, for example:

daemons(url = "tcp://:0", dispatcher = FALSE)
#> [1] "tcp://:37421"

Note that above, the port number is specified as zero. This is a wildcard value that will automatically cause a free ephemeral port to be assigned. The actual assigned port is provided as the return value of the call, or it may be queried at any time by requesting the status via daemons().

–

On the server, server() may be called from an R session, or an Rscript invocation from a shell. This sets up a remote daemon process that connects to the client URL and receives tasks:

Rscript -e 'mirai::server("tcp://10.111.5.13:37421")'

As before, daemons() should be set up on the client before launching server() on remote resources, otherwise the server instances will exit if a connection is not immediately available. Alternatievly specifying server(asyncdial = TRUE) will allow servers to wait (indefinitely) for a connection to become available.

–

The number of daemons connecting to the client URL is not limited and network resources may be added or removed at any time, with tasks automatically distributed to all server processes.

$connections will show the actual number of connected server instances.

daemons()
#> $connections
#> [1] 0
#> 
#> $daemons
#> [1] "tcp://:37421"

To reset all connections and revert to default behaviour:

daemons(0)
#> [1] 0

This causes all connected server instances to exit automatically.

« Back to ToC

Compute Profiles

The daemons() interface also allows the specification of compute profiles for managing tasks with heterogeneous compute requirements:

send tasks to different servers or server clusters with the appropriate specifications (in terms of CPUs / memory / GPU / accelerators etc.)
split tasks between local and remote computation

Simply specify the argument .compute when calling daemons() with a profile name (which is ‘default’ for the default profile). The daemons settings are saved under the named profile.

To launch a ‘mirai’ task using a specific compute profile, specify the ‘.compute’ argument to mirai(), which defaults to the ‘default’ compute profile.

« Back to ToC

Errors, Interrupts and Timeouts

If execution in a mirai fails, the error message is returned as a character string of class ‘miraiError’ and ‘errorValue’ to facilitate debugging. is_mirai_error() can be used to test for mirai execution errors.

m1 <- mirai(stop("occurred with a custom message", call. = FALSE))
call_mirai(m1)$data
#> 'miraiError' chr Error: occurred with a custom message

m2 <- mirai(mirai::mirai())
call_mirai(m2)$data
#> 'miraiError' chr Error in mirai::mirai(): missing expression, perhaps wrap in {}?

is_mirai_error(m2$data)
#> [1] TRUE
is_error_value(m2$data)
#> [1] TRUE

If during a call_mirai() an interrupt e.g. ctrl+c is sent, the mirai will resolve to an empty character string of class ‘miraiInterrupt’ and ‘errorValue’. is_mirai_interrupt() may be used to test for such interrupts.

is_mirai_interrupt(m2$data)
#> [1] FALSE

If execution of a mirai surpasses the timeout set via the ‘.timeout’ argument, the mirai will resolve to an ‘errorValue’. This can, amongst other things, guard against mirai processes that hang and never return.

m3 <- mirai(nanonext::msleep(1000), .timeout = 500)
call_mirai(m3)$data
#> 'errorValue' int 5 | Timed out

is_mirai_error(m3$data)
#> [1] FALSE
is_mirai_interrupt(m3$data)
#> [1] FALSE
is_error_value(m3$data)
#> [1] TRUE

is_error_value() tests for all mirai execution errors, user interrupts and timeouts.

« Back to ToC

Deferred Evaluation Pipe

mirai implements a deferred evaluation pipe %>>% for working with potentially unresolved values.

Pipe a ‘mirai’ or mirai $data value forward into a function or series of functions and it initially returns an ‘unresolvedExpr’.

The result may be queried at $data, which will return another ‘unresolvedExpr’ whilst unresolved. However when the original value resolves, the ‘unresolvedExpr’ will simultaneously resolve into a ‘resolvedExpr’, for which the evaluated result will be available at $data.

A piped expression should be wrapped in .() to ensure that the return value is always an ‘unresolvedExpr’ or ‘resolvedExpr’ as the case may be.

It is possible to use unresolved() around an expression, or its $data element, to test for resolution, as in the example below.

The pipe operator semantics are similar to R’s base pipe |>:

x %>>% f is equivalent to f(x)
x %>>% f() is equivalent to f(x)
x %>>% f(y) is equivalent to f(x, y)

m <- mirai({nanonext::msleep(500); 1})
b <- .(m %>>% c(2, 3) %>>% as.character)

unresolved(b)
#> [1] TRUE
b
#> < unresolvedExpr >
#>  - $data to query resolution
b$data
#> < unresolvedExpr >
#>  - $data to query resolution

nanonext::msleep(1000)
unresolved(b)
#> [1] FALSE
b
#> < resolvedExpr: $data >
b$data
#> [1] "1" "2" "3"

« Back to ToC

Integrations with Crew, Targets, Shiny

The crew package is a distributed worker-launcher that provides an R6-based interface extending mirai to different distributed computing platforms, from traditional clusters to cloud services. The crew.cluster package enables mirai-based workflows on traditional high-performance computing clusters such as Sun Grid Engine (SGE).

targets, a Make-like pipeline tool for statistics and data science, has integrated and adopted crew as its predominant high-performance computing backend.

mirai can also serve as the backend for enterprise asynchronous shiny applications in one of two ways:

mirai.promises, which enables a ‘mirai’ to be used interchangeably with a ‘promise’ in shiny or plumber pipelines; or
crew provides an interface that makes it easy to deploy mirai for shiny. The package provides a Shiny vignette with tutorial and sample code for this purpose.

« Back to ToC

Thanks

William Landau has been instrumental in shaping development of the package, from being the first to request persistent daemons, through to robustness testing for the high performance computing requirements of crew and targets.

Henrik Bengtsson has shared valuable insights leading to the interface accepting broader usage patterns.

« Back to ToC