I have been trying to calculate SVs using the sva-package in parallel, but no matter what backend I use, it always fails. Consider for examples this example code based on the vignette:

library(parallel)
library(BiocParallel)
library(sva)
library(bladderbatch)
data(bladderdata)

# Data from the vignette
pheno <- pData(bladderEset)
edata <- exprs(bladderEset)
mod <- model.matrix(~as.factor(cancer), data=pheno)

# Serial execution (Works)
svaSerial <- lapply(1:3, function(x) sva(dat=edata, mod=mod, n.sv=x))

# Parallel execution using parallel (Fails)
svaSerial <- mclapply(1:3, function(x) sva(dat=edata, mod=mod, n.sv=x), mc.cores=3)

# Parallel execution using BioC (Fails)
svaSerial <- bplapply(1:5, function(x) sva(dat=edata, mod=mod, n.sv=x), BPPARAM=MulticoreParam(workers=3))

 The code executes just fine under lapply, but mclapply, bpapply, etc. fails, without any informative error messages.

Any ideas why this is?

A simpler example might be

bplapply(1:2, sqrt, BPPARAM=MulticoreParam(2))

In BiocParallel version 1.8.1 or later you might try

options(bphost="localhost")
bplapply(1:2, sqrt, BPPARAM=MulticoreParam(2))

There might also be an issue with blocked ports. BiocParallel chooses a random port between 11000 and 12000. If these are blocked, specify an open port with

options(port=12345)