I would like to test out our own T2 site, and run at other sites, by using CRAB.  Now, how hard should this be, really?  Most of the documentation, I should note, makes it sound like the hard thing is taking care of your grid certificate.  Fortunately, I already have one of those.

One problem with CRAB is that apparently right now only SL3 systems can be use to submit the jobs.  What a pain!  It's a known serious problem.  So I moved over to cmsuaf.fnal.gov, which is an SL3 system.  I successfully installed my grid certificate there (see "Grid stuff") and was able to get a proxy, etc.  Actually, let me note right here that for this purpose, the command you must issue is

voms-proxy-init cms

This gives you credentials in the right VO.  I had to enter my passphrase (I'll have to keep remembering that!), and then I was granted a proxy that's good for twelve hours.  And of course I also had to re-establish my analysis code over at cmsuaf; I did that by making a new project, and doing a new mkedanlzr (gosh, that's a terrible name), and then I copied code and config file from gpn-husker before scramv1 build etc.  All seems right with that.

As for CRAB itself -- I would be dead in the water with out Oliver Gutsche, who wrote some excellent step-by-step instructions for how to use CRAB.  Follow that, for sure!  (It's definitely better than what I found on the CRAB documentation pages.)  But it covers many things that I already knew how to do.  On top of that, it also has a lot of instructions on how to install CRAB on your own machine, which isn't necessary on cmsuaf.  (We should make it so on gpn-husker too -- put the CRAB stuff in some reasonable directory.)  So the new stuff is:

Set up the CRAB stuff:

source /uscmst1/prod/grid/CRAB_1_5_1/crab.sh

(Oh look, I copied it from Oliver's twiki page, and I get that lovely formatting!)  You can put in a different version number, of course.  The very first time you do this, you need to set stuff up for BOSS:

$CRABDIR/configureBoss

I stumbled here the first time because I already had a .bossrc area in my home directory.  Make sure that you don't.

Then I needed a crab.cfg file.  This is the part that looks most dangerous, as there are many parameters that need to be set.  To first order, you could probably follow what is done in Oliver's example file.  And you could also get some help from the DBS discovery page.  When you find a dataset, you can click on the "more" link next to the dataset, and get a crab.cfg.  Within that, I had to set my "pset" and "output_file" parameters, and specify "outputdir" and "logdir".  I'll also note that Oliver's version uses the edg scheduler, while DBS pointed me to glitecoll.  I chose the latter, at least for now.  And I'm using the grid middleware sandbox for my output return.  Since I'm only creating a histogram file, this should be quite reasonable.  But if I were outputting any data file at all, I'd probably need to figure out how to use a storage element.

Anyhow, if I actually get all of that right, the next steps should be easy.  First, I create a job:

crab -create

For the particular dataset I chose, and the requirement of 1000 events/job, it made four jobs for submission.  And then I try submitting those jobs:

crab -submit all -c

I think the point of the -c is that you might have created multiple (sets of) CRAB jobs, and this points you to the most recent one.  (The directories from earlier attempts hang around.) 

But this is where I'm currently stuck.  The jobs fail, with a complaint that there's some issue with the CMSSW version I want not being available at the sites.  How could that be?  You would figure that this would be figured out when the job is created.  I'm waiting for more feedback from Oliver on this.