CODEWITHSUNDEEP
rforeachdoparallel
Philip Swannell
Philip Swannell

Reputation: 935

foreach %dopar% yielding “cannot find function” errors when the function to be parallelised is a list element

I’ve successfully parallelised a function – let’s call it AddOne - via the doParallel package, foreach and %dopar% and I’m familiar with the .packages and .export arguments to foreach.

My problem is that I would like AddOne, instead of being a “stand-alone” function, to be an element of a list and in this case, I can’t get things working. Specifically, if AddOne calls a subroutine AddOneSubroutine then AddOneSubroutine does not get found in the “worker” environments even though it is “exported”.

I’m using Windows 10 and R.version yields:

platform       x86_64-w64-mingw32          
arch           x86_64                      
os             mingw32                     
system         x86_64, mingw32             
status                                     
major          3                           
minor          4.1                         
year           2017                        
month          06                          
day            30                          
svn rev        72865                       
language       R                           
version.string R version 3.4.1 (2017-06-30)
nickname       Single Candle

The doParallel version I have is 1.0.10. Here’s some code that demonstrates the problem as succinctly as I could.

library(doParallel)
if(!exists("Registered")){
    registerDoParallel(cores = detectCores(logical = TRUE))
    Registered = TRUE
}

AddOne<-function(x){AddOneSubroutine(x)}
AddOneSubroutine <-function(x){x+1}

MyList<-list()
MyList$f<-AddOne

# Not using parallel environments, works correctly when calling AddOne 3 times
Result1 = foreach(i = 1:3) %do% AddOne(i)
Result1

# Not using parallel environments, works correctly when calling MyList$f 3 times
Result2 = foreach(i = 1:3) %do% MyList$f(i)
Result2

# Using parallel environments, works correctly when calling AddOne 3 times,
# despite not explicitly using the .export argument to export AddOneSubroutine
Result3 = foreach(i = 1:3) %dopar% AddOne(i)
Result3

# Using parallel environments, fails when calling MyList$f with error
# "could not find function "AddOneSubroutine"", even though that function is "exported"
Result4 = foreach(i = 1:3,.export = "AddOneSubroutine") %dopar% MyList$f(i)
Result4

What am I failing to understand?

Upvotes: 0

Views: 2191

Answers (1)

HenrikB
HenrikB

Reputation: 6815

For full reproducibility everywhere, let us make sure we use workers in background R sessions:

library("doParallel")
cl <- parallel::makeCluster(detectCores(logical = TRUE))
registerDoParallel(cl)

Now, I haven't dug into the doParallel backend code in much detail, so I'm not 100% sure what causes this problem. But we know that AddOneSubroutine is indeed exported, which you can see if you use foreach(..., .verbose = TRUE), or simply do:

AddOneSubroutine <- function(x) { x + 1 }
y <- foreach(i = 1L, .export = "AddOneSubroutine") %dopar% {
    get("AddOneSubroutine")
}
str(y)
## List of 1
##  $ :function (x)  
##   ..- attr(*, "srcref")=Class 'srcref'  atomic [1:8] 1 20 1 40 20 40 1 1
##   .. .. ..- attr(*, "srcfile")=Classes 'srcfilecopy', 'srcfile' <environment: 0x2e475a0>

However, when calling the function MyList$f() it is not found, which can be confirmed by using:

AddOne <- function(x) exists("AddOneSubroutine")
MyList <- list()
MyList$f <- AddOne
y <- foreach(i = 1L, .export = "AddOneSubroutine") %dopar% {
    MyList$f(i)
}
str(y)
## List of 1
##  $ : logi FALSE

So, why is AddOneSubroutine not in the frames searched from within MyList$f? This could be because doParallel does not get the environment for MyList$f correct. A workaround that seems to work, is the following hack:

AddOne <- function(x) { AddOneSubroutine(x) }
y <- foreach(i = 1L) %dopar% {
    environment(MyList$f) <- environment(AddOneSubroutine)
    MyList$f(i)
}
str(y)
## List of 1
##  $ : num 2

Unfortunately, it's not very neat nor very convenient.

As an alternative, the doFuture backend (I'm the author) seems to work a slightly better:

library("doFuture")
registerDoFuture()
plan(multisession)

AddOneSubroutine <- function(x) { x + 1 }
AddOne <- function(x) { AddOneSubroutine(x) }
MyList <- list()
MyList$f <- AddOne

y <- foreach(i = 1L) %dopar% {
    AddOneSubroutine ## dummy guiding auto-export
    MyList$f(i)
}
str(y)
## List of 1
##  $ : num 2

PS. You're particular use case interested me because ideally AddOneSubroutine should have been exported automatically when using doFuture but it didn't. I've found a fix for this in the underlying globals package (I'm the author) but I need to think more about it before publishing it.

My details:

> sessionInfo()
R version 3.4.1 (2017-06-30)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 16.04.3 LTS

Matrix products: default
BLAS: /usr/lib/atlas-base/atlas/libblas.so.3.0
LAPACK: /usr/lib/atlas-base/atlas/liblapack.so.3.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] parallel  stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
[1] doFuture_0.5.1  iterators_1.0.8 foreach_1.4.3   future_1.6.1   

loaded via a namespace (and not attached):
[1] compiler_3.4.1   tools_3.4.1      listenv_0.6.0    codetools_0.2-15
[5] digest_0.6.12    globals_0.10.2

Upvotes: 3

Related Questions