R/buildCluster.R

In bogdanoancea/aggregation: R package for aggregation of device numbers into population number

#' @title Builds a cluster of working nodes.
#'
#' @description Builds a cluster of working nodes using the \code{parallel}
#'   package. On Unix-like operating systems it builds as "FORK" cluster while
#'   on Windows it builds a "SOCKET" cluster. The number of working nodes equals
#'   the number of available (logical) cores as it is detected by the call of
#'   \code{detectCores()} function. On Windows, it also exports a list of
#'   variables and an environment passed as arguments. This is an internal
#'   function of the package, and it is not available to be direclty called by
#'   users.
#'
#' @param varlist A list of variables to be exported to every working node.
#'
#' @param env An environment to be exported to every working node.
#'
#' @return A cluster object used to distribute the computations.
#'
#' @keywords internal
#' @import data.table
#' @import extraDistr
#' @import parallel
#' @import doParallel
#' @export
buildCluster <- function(varlist, env) {

  if (Sys.info()[['sysname']] == 'Linux' |
      Sys.info()[['sysname']] == 'Darwin') {
    cl <- makeCluster(detectCores(), type = "FORK")
  } else {
    cl <- makeCluster(detectCores())
    clusterEvalQ(cl, library("data.table"))
    clusterEvalQ(cl, library("extraDistr"))
    clusterEvalQ(cl, library("deduplication"))
    clusterEvalQ(cl, library("Matrix"))
    clusterExport(cl, varlist, envir = env)
  }
  return (cl)
}