Scope of variables in parallel R package snowfall

Question

As my understanding, the slave processes cannot access variables in the global/parent environments in R package snowfall.

For example, the code below will give an error message as I expected:

parfun <- function(i, var1, var2) {
    return(var3)
}

var3 <- 'VAR3'
library(snowfall)
sfInit(parallel = TRUE, cpus = 2)
res <- sfLapply(seq(1, 2), parfun, 'VAR1', 'VAR2')
res

The error message:

Error in checkForRemoteErrors(val) : 
  2 nodes produced errors; first error: object 'var3' not found

However, there is no error message when parallel codes in another function.

fun1 <- function(var1, var2, var3) {
    parfun <- function(i, var1, var2) {
        return(var3)
    }
    library(snowfall)
    sfInit(parallel = TRUE, cpus = 2)
    res <- sfLapply(seq(1, 2), parfun, var1, var2)
    res
}

fun1('VAR1', 'VAR2', 'VAR3')

The results are as follows:

[[1]]
[1] "VAR3"

[[2]]
[1] "VAR3"

I don't understand this behavior as var3 doesn't exist in the function parfun, but in his parent environment fun1.

Could somebody explain to me? Thanks for any suggestions.

Scope of variables in parallel R package snowfall

0 Answers0