使用parallel函数时R CRAN检查失败的问题求助

The Problem #

You're trying to submit an R package to CRAN that uses parallel::makeCluster(parallel::detectCores()) for parallel computation. Local builds work fine, but devtools::check(document = FALSE) throws an error:

Running examples in ‘TESTER-Ex.R’ failed
The error most likely occurred in:
> base::assign(".ptime", proc.time(), pos = "CheckExEnv")
> ### Name: hello_world
> ### Title: Prints hello world
> ### Aliases: hello_world
> 
> ### ** Examples
> 
> hello_world()
Error in .check_ncores(length(names)) : 8 simultaneous processes spawned
Calls: hello_world -> <Anonymous> -> makePSOCKcluster -> .check_ncores
Execution halted

You reproduced this with a minimal package containing just the hello_world function:

#' Prints hello world
#'
#' @return nothing
#' @export
#'
#' @examples
#' hello_world()
hello_world <- function() {
  # initiate cluster
  cl <- parallel::makeCluster(parallel::detectCores())
  # stop cluster
  parallel::stopCluster(cl)
  cat("Hello World\n")
  return(invisible(NULL))
}

Why This Happens #

This is a resource restriction enforced by CRAN's check servers. CRAN sets the environment variable _R_CHECK_LIMIT_CORES_ to limit the number of parallel processes packages can spawn (usually to 2 cores) — this prevents packages from hogging shared server resources.

When you call parallel::detectCores(), it returns the total number of cores available on the machine (in your error case, 8). This exceeds the CRAN-imposed limit, triggering the .check_ncores() validation inside makePSOCKcluster() which throws the error.

Your local environment doesn't have this restriction, which is why everything works fine locally but fails during CRAN checks.

How to Fix It #

Adjust your code to respect the CRAN core limit when it's set, and fall back to a reasonable default otherwise. Here's how to modify your hello_world function:

#' Prints hello world
#'
#' @return nothing
#' @export
#'
#' @examples
#' hello_world()
hello_world <- function() {
  # Determine safe number of cores to use
  n_cores <- if (Sys.getenv("_R_CHECK_LIMIT_CORES_") != "") {
    # Use CRAN's allowed core count if set
    as.integer(Sys.getenv("_R_CHECK_LIMIT_CORES_"))
  } else {
    # Fall back to physical cores (or 2 if detection fails)
    nc <- parallel::detectCores(logical = FALSE)
    if (is.null(nc)) 2 else nc
  }
  
  # Initiate cluster with the safe core count
  cl <- parallel::makeCluster(n_cores)
  # Stop cluster
  parallel::stopCluster(cl)
  cat("Hello World\n")
  return(invisible(NULL))
}

Key Notes:

• We check for the _R_CHECK_LIMIT_CORES_ environment variable first — this ensures we comply with CRAN's rules during checks.
• Using detectCores(logical = FALSE) gives us physical cores instead of logical (hyper-threaded) cores, which is often more appropriate for parallel work.
• We add a fallback to 2 cores if core detection fails (rare, but handles edge cases).

Alternatively, if your parallel code isn't critical to demonstrate in the example, you could wrap the parallel logic in \dontrun{} or \donttest{} in your roxygen comments. But the core-count adjustment is the better long-term solution, as it keeps your example runnable while respecting CRAN's constraints.

内容的提问来源于stack exchange，提问作者David