Error: there is no package called ‘doFuture’ when run on cluster

[KatieBuc]

Hello,

I am running an r script which works locally, looks like:

library(foreach)
library(parallel)
library(doFuture)
...
doFuture::registerDoFuture()
cl <- parallel::makeCluster(10)
plan(cluster, workers = cl)
...
utility <- foreach(i = 1:5, .combine = c, .packages = c("SSN", "SSNdesign", "mvtnorm", "doFuture", "parallel")) %dopar% {
...}

Which works fine locally but when run on the cluster gives,

Error: there is no package called ‘doFuture’

only after the foreach is called, not on the library import.

I've tried various other calls like

#cl <- parallel::makeCluster(10, libs = .libPaths('//home//n10907700//packages'))
#cl <- future::makeClusterPSOCK(1L, verbose=TRUE)
#cl <- parallel::makeCluster(1L, rscript=file.path(R.home("bin"), "Rscript"))

and also

#plan(cluster, workers = cl, rscript_libs = .libPaths('//home//n10907700//packages'))

same error... when I tried running sequentially on the cluster it works

plan(sequential)

So I'm thinking it's a futures framework thing?

Here's some more info

sessionInfo()
R version 3.6.2 (2019-12-12)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: SUSE Linux Enterprise Server 12 SP5
Matrix products: default
BLAS/LAPACK: /pkg/suse12/software/OpenBLAS/0.3.7-GCC-8.3.0/lib/libopenblas_haswellp-r0.3.7.so
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] stats graphics grDevices utils datasets methods base
loaded via a namespace (and not attached):
[1] compiler_3.6.2

also,

print(.libPaths())
[1] "/home/n10907700/packages"
[2] "/pkg/suse12/software/R/3.6.2-foss-2019b/lib64/R/library"
print(find.package("doFuture"))
[1] "/home/n10907700/packages/doFuture"

and

R.home()
[1] "/pkg/suse12/software/R/3.6.2-foss-2019b/lib64/R"
parallel::clusterEvalQ(cl, R.home())
[[1]]
[1] "/pkg/suse12/software/R/3.6.2-foss-2019b/lib64/R"
[[2]]
[1] "/pkg/suse12/software/R/3.6.2-foss-2019b/lib64/R"
[[3]]
[1] "/pkg/suse12/software/R/3.6.2-foss-2019b/lib64/R"
...

Any ideas? Thank you!

[HenrikBengtsson]

Hi. Before anything else, what's your sessionInfo() after the error occurs? It'll show exactly what packages and versions are involved.

I suspect that this has to do with you having a customized .libPaths() in your R startup and the solution is to make sure that is also inherited by the workers. To see what the first worker sees, do:

doFuture::registerDoFuture()
paths <- foreach(x = 1) %dopar% { .libPaths() }
print(paths)

[KatieBuc]

Hi Henrik,

Thanks for your quick reply. I've now manually installed the packages on the cluster, it's set up the lib destination automatically and I have not done anything with .libpaths() this time around. There is no error (yay!), but I'm back to the original problem, which is that it's hanging on the cluster.

This R script takes 33 seconds to run locally, whereas on the cluster I killed it after 2 mins and it made it through just 1 of 30 iterations that had otherwise locally completed in that time. Is this a common issue? Is there a fix?

For what it's worth the info:

sessionInfo()
R version 3.6.2 (2019-12-12)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: SUSE Linux Enterprise Server 12 SP5
Matrix products: default
BLAS/LAPACK: /pkg/suse12/software/OpenBLAS/0.3.7-GCC-8.3.0/lib/libopenblas_haswellp-r0.3.7.so
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] stats graphics grDevices utils datasets methods base
loaded via a namespace (and not attached):
[1] compiler_3.6.2

print(paths)
[1] "/pkg/suse12/software/rgdal/1.4-8-foss-2019b-R-3.6.2"
[2] "/home/n10907700/R/x86_64-pc-linux-gnu-library/3.6"
[3] "/pkg/suse12/software/R/3.6.2-foss-2019b/lib64/R/library"

Thank you,
Katie

[HenrikBengtsson]

I'd start by verifying that parallelization actually works with your cluster setup, e.g.

library(doFuture)
registerDoFuture()
cl <- ...   ## <= I don't what you do here; it may be important
plan(cluster, workers = cl)
info <- foreach(x = 1:nbrOfWorkers(), .combine = rbind) %dopar% { 
  data.frame(hostname = Sys.info()[["nodename"]], rversion = getRversion(), pid = Sys.getpid())
}
print(info)

[KatieBuc]

library(doFuture)
library(parallel)
registerDoFuture()
cl <- parallel::makeCluster(10)
plan(cluster, workers = cl)
info <- foreach(x = 1:nbrOfWorkers(), .combine = rbind) %dopar% {
data.frame(hostname = Sys.info()[["nodename"]], rversion = getRversion(), pid = Sys.getpid())
}
print(info)
hostname rversion pid
1 cl4n010 3.6.2 45887
2 cl4n010 3.6.2 45900
3 cl4n010 3.6.2 45913
4 cl4n010 3.6.2 45926
5 cl4n010 3.6.2 45939
6 cl4n010 3.6.2 45952
7 cl4n010 3.6.2 45965
8 cl4n010 3.6.2 45978
9 cl4n010 3.6.2 45991
10 cl4n010 3.6.2 46004

Hmm what's the verdict?

[HenrikBengtsson]

Sorry for the delay. Your output confirms that the basic setup of 10 workers works as it should. That's good and helps rule out certain things. I realized my comment in #490 (comment) was ambiguous - I wasn't explicit about the plan() - it might be that your reply was not what I needed to know. So, what does:

library(doFuture)
library(parallel)
registerDoFuture()
cl <- parallel::makeCluster(2)
plan(cluster, workers = cl)
paths <- foreach(x = 1L) %dopar% { .libPaths() }
print(paths)

output?

I suspect that your troubles has to do with the fact that you've got a peculiar .libPaths() setup; it's not the default that R would give you and I depending on how and where your is configured, it might cause problems for your parallel workers. So, I'm asking "troubleshooting" questions to narrow in on this.