README.md

In erblast/updateR: Update R in a desktop and server environment

updateR

provides workflow for updating R packages and maintaining a miniCRAN repository in an offline server environment

Introduction

In a corporate IT environment handling R and R package updates can be tricky. Usually R packages are installed into the user's HOME directory while standard library packages can be found inside the R installation's library directory. These paths can be accessed via .libPaths(). The R isntallation itself can usually be found under ./Program Files

When updating to a new R version only the packages inside the library directory of the R installation will be replaced while all packages inside the HOME directory remain unchanged. Usually after installing a new version of R we will also update all packages using update.packages().

Restrictions / Requirements of a corporate IT landscape

1. We might have limited access to the HOME and the ./Program Files directories for IT security reasons
2. We can only access the local database from a server without internet access, thus cannot install packages from online sources
3. The code needs to be reproducible, thus we need to archive older versions of R and their packages.

How to set up R to be compliant

1. Do not install packages into the HOME directory and install R to a local drive outside ./Program Files. You can setup up R to only install into the R installation's library directory by adding libPaths( <path> ) to the Rprofile.site file inside the etc folder of the R installation's parent directory.
2. We can use the miniCRAN package to create a local CRAN repository on a machine with online connection which we can then copy to the server.
3. We can use packages like packrat or checkpoint they will save all packages for a certain analysis on disk. This approach uses up a lot of disk space and can be quite slow. Another approach could be to use a docker image. What I find more practible is to create a snapshot everytime R is updated to a new version, which is approximately 4 times a year. Before installing the new version we simply archive the old R version and the old version of the miniCRAN and note in each analysis folder which R version was used. If R code is not running anymore on the current version we can simply spin up the old one with which the analysis was orignally performed. This is not a 100% fool-proofed like saving all packages to disk for each analysis but is more disk space friendly.

Callenges of the proposed workflow

This workflow requires a lot of copy and pasting of files and it is easy to make mistakes.

• running update.packages(), packages that are explicitly or implicitly loaded will fail to update and then be removed from the current library. There are a number of important packages which are always loaded when Rstudio is running like Rccp, rlang to just name a few. Also another R Session running will cause problems.

• Each of the steps that are needed to update to a new version takes quite a long time which makes it easy to forget on which step of the sequence one currently is.

How to use updateR

• updateR is MS Windows only
• make sure no packages are installed in the HOME directory
• install updateR for old Rversion on server and desktop
• download and install new R version on server and desktop
• do not use Rstudio to run any of the updateR functions
• for example run <R installation path>/bin/<x64 or i386>/Rgui.exe instead.

update R will perform a number of sequential steps for you with numerous checks and controls so everything goes smoothly and as fast as possible.

When updating to a new R version

Note all of these function will aks you for some user input - Select new and old R installation folder - Select whether you are on a server or desktop - Select whether you are using miniCRAN on the desktop

Create miniCRAN

run in old R Version

If you do not have a miniCRAN yet make one.

updateR::create_miniCRAN( path = "c:/miniCRAN" )

Update new installation from old installation on desktop

run in old R Version

updateR::update_from_old_inst()

• copies all libraries from old to new version
• updates Rprofile.site

Update new installation from old installation on server

run in old R Version

updateR::update_from_old_inst()

• copies all libraries from old to new version
• updates Rprofile.site
• archives miniCRAN

Update new installation from CRAN repository on desktop

run in new R Version

updateR::update_new_inst()

• this will update all R packages
• will add any local packages available on CRAN but not miniCRAN to miniCRAN
• update miniCRAN

Transfer miniCRAN from desktop to server

Before you transfer make sure that you ran updateR::update_new_inst() on server first and check if you can find the archived miniCRAN version on disk.

Note this is the only step where you can make a mistake. You will end up with one archived miniCRAN less

This step is important for updating the packages on the server and installing new packages.

Update new installation from CRAN repository on server

run in new R Version

updateR::update_new_inst()

• this will install all packages found on miniCRAN which are not yet installed locally
• will update all packages from miniCRAN

Install new CRAN packages on the server

When prompted to select the directories of the R installations simply select the most recent installation twice.

• install them on desktop via install.packages
• run updateR::update_new_inst() on desktop
• transfer miniCRAN from desktop to server
• run updateR::update_new_inst() on server

Note, I would not archive the miniCRAN between R releases, you can however do so manually updateR::update_from_old_inst() will not work if there is alrady an archive named after the current R Version.

Technical Notes

• updateR only uses base R and miniCRAN functions
• updateR does not regularly import miniCRAN because miniCRAN implicitly loads igraph which then can fail to update properly. We therefore load miniCRAN via require('miniCRAN') and then unload the package via detach( 'package:miniCRAN', unload = T)
• this causes updateR to produce warnings during package checking

erblast/updateR documentation built on May 17, 2019, 7:04 p.m.