Segfault when reading CSV from within the R environment

[Tmonster]

I am in the process of setting up a new run the the db-benchmark and Julia keeps segfaulting when reading the CSV for the data. Maybe I'm just holding it wrong, but the presence of a segfault and not an error tells me there is an actual bug.

This is the most minimum working example I could make

files needed

julia_segfault.R

system("julia julia_script.jl")

julia_script.jl

#!/usr/bin/env julia

using InMemoryDatasets;
using Printf;
using DLMReader;
using PooledArrays;

x = filereader("small.csv", types=[Characters{5}, Characters{5}, Characters{12}, Int32, Int32, Int32, Int32, Int32, Float64]);

print("done loading...\n"); flush(stdout);

CSV file
small.csv

Command to reproduce and error

$ Rscript julia_segfault.R
Segmentation fault (core dumped)

Originally I thought this could be an R issue, but since the Julia script never finishes, I think the problem is more on the Julia side.

versioninfo()

Julia Version 1.10.5
Commit 6f3fdf7b362 (2024-08-27 14:19 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 32 × Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, icelake-server)
Threads: 40 default, 0 interactive, 20 GC (on 32 virtual cores)
Environment:
  JULIA_HOME = /opt/julia-1.10.5
  JULIA_NUM_THREADS = 40

R sessioninfo()

R version 4.3.3 (2024-02-29)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 24.04.1 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.12.0
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.12.0

locale:
 [1] LC_CTYPE=C.UTF-8       LC_NUMERIC=C           LC_TIME=C.UTF-8
 [4] LC_COLLATE=C.UTF-8     LC_MONETARY=C.UTF-8    LC_MESSAGES=C.UTF-8
 [7] LC_PAPER=C.UTF-8       LC_NAME=C              LC_ADDRESS=C
[10] LC_TELEPHONE=C         LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C

time zone: Etc/UTC
tzcode source: system (glibc)

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

loaded via a namespace (and not attached):
[1] compiler_4.3.3

Let me know if you need anymore information, this was reproduced on an aws machine with size c6id.8xlarge

Edit: Added error message

[oscardssmith]

Do you have any LD_PRELOADS that could be forcing Julia to load incorrect versions of libraries?

[giordano]

Let me know if you need anymore information

For example the full error message, I don't see it anywhere, or am I totally missing it?

[Tmonster]

Do you have any LD_PRELOADS that could be forcing Julia to load incorrect versions of libraries?

Not too familiar with LD_PRELOADs, so I am going to say no. The aws machine I used was fresh and used fresh installs of the libraries, so I don't think there are any incorrect versions being loaded, unless the libraries I'm using aren't available for 1.10.5.

For example the full error message, I don't see it anywhere, or am I totally missing it?

I forgot to add it, but it isn't very interesting

$ Rscript julia_segfault.R
Segmentation fault (core dumped)

[xgdgsc]

What if you call a bash script that calls julia in system within R?

[bkamins]

CC @quinnj

[ViralBShah]

I don't see a segfault on 1.10.7 or 1.11.2, and the dataset loads just fine. I'm using mac arm64. I ran it standalone and from R 4.4.2:

> system("julia julia_script.jl")
┌ Warning: Julia started with single thread, to enable multithreaded functionalities in InMemoryDatasets.jl start Julia with multiple threads.
└ @ InMemoryDatasets ~/.julia/packages/InMemoryDatasets/5VOs4/src/InMemoryDatasets.jl:208
┌ Warning: the last line is not ended with `end of line` character
└ @ DLMReader ~/.julia/packages/DLMReader/ID9gW/src/reader.jl:449
done loading...

[ViralBShah]

Let's reopen if we can have a reproducer.

[Am386DX-40]

I can reproduce it for Ubuntu 24.04.01 LTS (on WSL2 on Win11) with r-base 4.3.3 package installed from the Ubuntu package repository.

The MWE on Julia 1.10.7 LTS crashes with gdb backtrace gdb.txt

R sets the LD_LIBRARY_PATH:

$ echo $LD_LIBRARY_PATH

$ R --silent -e "Sys.getenv('LD_LIBRARY_PATH')"
> Sys.getenv('LD_LIBRARY_PATH')
[1] "/usr/lib/R/lib:/usr/lib/x86_64-linux-gnu:/usr/lib/jvm/default-java/lib/server"
>
>

Using system("LD_LIBRARY_PATH='' julia julia_script.jl") in julia_segfault.R solves the issue.

The MWE doesn't produce a segfault on Julia 1.11.2.

[ViralBShah]

There seem to be issues running R and Julia together on Ubuntu 24: JuliaInterop/JuliaCall#238. I wonder where the actual issue is.

[nalimilan]

Looks like the explanation has been found: JuliaInterop/JuliaCall#238 (comment)

[nalimilan]

@Tmonster According to JuliaInterop/JuliaCall#238 (comment), the problem comes from the wrapper Ubuntu uses to run R: it sets LD_LIBRARY_PATH, which is a hack which is expected to break other programs run from the same environment. In the present case, an incorrect version of libunwind is loaded instead of the one Julia ships with.

It might be enough to call Sys.setenv(LD_LIBRARY_PATH="") before calling system("julia julia_script.jl"), or maybe better call system2("julia julia_script.jl", env="LD_LIBRARY_PATH="). Would you mind trying that? It's too bad the benchmark is broken for Julia just because of this bad interaction with R in Ubuntu.