RStudio server¶

Hydra has a dedicated RStudio server at https://galaxy.si.edu/R4. A browser opens an RStudio session on a node with 192 CPU cores and 1.5 TB of memory, with your Hydra home directory, /data, /scratch and /store mounted. It runs R 4.5.3 and RStudio Server 2024.12.0+467, the same for every user, and needs no tunnel and no qrsh.

Log in¶

From the SI network or the SI VPN, open https://galaxy.si.edu/R4 in the browser:

From telework.si.edu, enter https://galaxy.si.edu/R4 in the Enter an internal resource box on the telework home page and press Return:
Sign in with your Hydra username, in lower case, and your Hydra password:
RStudio opens in the browser with the same layout as the desktop version:

The Files pane shows Hydra's partitions. Navigate to a directory under /scratch or /data to work there:

The Files pane showing a directory under /scratch

Sessions¶

The node is shared with every other user. One R session per user runs at a time. Logging in from a second browser takes over the existing session rather than starting another. A session keeps running when you close the browser or log off your computer, with its objects in memory and its computations under way. Logging in again reconnects to it. To run several analyses at once inside the one session, use the Background Jobs tab (Posit's guide).

Quit the session when you are done

An idle session holds memory that other users cannot get. Session and then Quit Session… ends it. Closing the browser does not.

The Quit Session item in the Session menu

If the server refuses to log you back in after you used its log-off button, clear the browser's cookies for galaxy.si.edu.

Move files in and out¶

The Files pane transfers files. For anything large or many files, use Data transfer instead.

Upload: Upload in the Files pane, one file at a time. To upload several, zip them on your computer. The server unzips the archive on arrival.
Download: tick the files or folders, then More and Export…. Several files or a folder come down as one zip.

Install packages¶

install.packages() and BiocManager::install() work as on a desktop, into your personal library at ~/R/x86_64-redhat-linux-gnu-library/4.5, which other users do not see. Every package compiles from source on Linux, so installs take minutes. Ncpus parallelises them:

install.packages("seqinr", Ncpus = 8)
BiocManager::install("phyloseq", Ncpus = 8)

Packages that need a newer GCC¶

Some packages (ade4 and seqinr among them) need a newer compiler than the server's system GCC and fail with missing GLIBCXX symbols or OpenMP errors. The server cannot load compiler modules, so point R at one of Hydra's GCC installations with a ~/.R/Makevars file.

If ~/.Renviron sets compiler paths, remove it. Mixed compiler settings are the usual cause of the errors:
```
$ rm ~/.Renviron
```

Create the Makevars file, in the RStudio Terminal (not the R console):

$ mkdir -p ~/.R
$ nano ~/.R/Makevars

~/.R/Makevars

CC = /share/apps/tools/gcc/14.2.0/bin/gcc
CXX = /share/apps/tools/gcc/14.2.0/bin/g++
CXX11 = /share/apps/tools/gcc/14.2.0/bin/g++
CXX14 = /share/apps/tools/gcc/14.2.0/bin/g++
CXX17 = /share/apps/tools/gcc/14.2.0/bin/g++
FC = /share/apps/tools/gcc/14.2.0/bin/gfortran
F77 = /share/apps/tools/gcc/14.2.0/bin/gfortran

CFLAGS = -O2 -g -fopenmp -fpic
CXXFLAGS = -O2 -g -fopenmp -fpic
CXX11FLAGS = -O2 -g -fopenmp -fpic
CXX14FLAGS = -O2 -g -fopenmp -fpic
CXX17FLAGS = -O2 -g -fopenmp -fpic
FFLAGS = -O2 -g -fpic
FCFLAGS = -O2 -g -fpic

Every line names the same GCC. 14.2.0 can be any of 9.2.0, 9.3.0, 10.1.0 (OpenMP 4.5), 11.2.0, 12.2.0 (OpenMP 5.0), or 13.2.0, 14.2.0, 15.2.0 (OpenMP 5.1). The ade4 package needs 10.1.0 or newer.

Session and then Restart R. The compiler settings take effect in the new session.
Install as usual.

To go back to the system GCC, delete ~/.R/Makevars and restart R. A package that installs but fails to load points at a Makevars file in which the lines name different GCC versions.

Temporary files¶

R writes temporary files under /tmp on the node, which is small and shared. Put them under your own space by setting TMPDIR in ~/.Renviron, or in .Renviron in the project directory. A shell export has no effect, because the server's R session does not read your shell's environment:

~/.Renviron

TMPDIR=/scratch/genomics/USERNAME/tmp