R updates

[Predrag Punosevac]

mille856 at andrew.cmu.edu wrote:

> Predrag,
>    It seems to me that when R is updated on the compute nodes, the
> libraries are not rolled into the new version. If you have time, could
> you look into automating the reinstallation of packages?
> 
> http://stackoverflow.com/questions/1401904/painless-way-to-install-a-new-version-of-r-on-windows
> 
> -Kyle

Dear Autonians,

Finally, I put some time into R installation and this is a brief summary
of what I found out:

1. We use official EPEL (Extra Packages for Enterprise Linux) repo for R
which is second to none. These packages are officially approved by Red
Hat corporation. 

2. I have enabled yum-cron on all computing nodes and desktops 

root at lov3$ service yum-cron status
Nightly yum update is enabled

Such aggressive updating practice sometimes lead to breakage like in
the case of plotting with R three weeks ago. Since we use high quality
repository packages are promptly fixed. You can check for yourself that
plotting in R is working on your desktops and computing nodes and that
you are running the latest R-3.2.2. We are research Laboratory and
having more conservative policy runs against my scientific instincts.


3. Some cutting-edge research grade packages are installed using built
in R package manager. We have no choice with those. Remember we are
research grade facility not a bank or insurance company. I am not 100%
sure if those get broken during minor version upgrades. They typically
get installed in

root at lov3$ pwd
/usr/lib64/R/library

I have run 

update.packages(checkBuilt=TRUE, ask=FALSE)

on LOV3 as a test. If you are heavy R user please try to use R on LOV3
and look for the signs of broken packages. Please report any new issues.


4. Some of you have installed your own R packages with your normal user
privileges as in Kyles case.

Those packages get installed in

[mille856 at line x86_64-redhat-linux-gnu-library]$ ls
3.0  3.1  3.2

Since Kyle is running R-3.2.2 now obviously the packages he installed
should be in 3.2. You can try to update only those using something like
grepping the list of outdated packages  which can be found running the
command 

> installed.packages()

and then selecting only your own packages from that list. If that sounds
too complicated you can just run 

update.packages(checkBuilt=TRUE, ask=FALSE)

which will upgrade all packages to the latest and the greatest version
including your own. However all those packages will be saved locally.
That is OK. I am OK having a multiple copies of R packages. There is
nothing I as a system admin can do about your custom installed R
packages. Once you install them you own them and you are in charge with
updates.

The link Kyle provided 

http://stackoverflow.com/questions/1401904/painless-way-to-install-a-new-version-of-r-on-windows

is very good and contains wealth of information. However not all that
info is applicable to our environment. If you need any assistance
processing the information from that web site please ping me.


Best,
Predrag