Bugs with toil

[suhrig]

Hello,

I tried running run_toil.sh on a LSF cluster, but ran into a few problems:

• I tried creating a conda environment using the file conda_toil_env.yml and Miniconda 4.9.2. The following packages could not be found. I had to remove the build numbers and in some cases even the version numbers to complete the installation:

Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound: 
  - nodejs==6.10.3=0
  - tk==8.5.18=0
  - readline==6.2=2
  - setuptools==36.4.0=py27_1

• Maybe you want to mention in the README that variables inside the run_toil.sh script need to be modified before running it, as this may not be obvious to users.
• The variable BATCHSYSTEM is defined but never used, which results in jobs being dispatched on the local machine instead of as LSF jobs. The variable should be passed to toil via the parameter --batchSystem.
• The variable JOBSTORE is undefined causing toil to attempt to create files in the root directory. This fails due to lack of permissions and the jobs crash.
• No runtime limits are configured for LSF jobs. On our cluster, jobs without explicit runtime specification get a default limit of just 10 minutes. Most jobs take longer than this and are terminated prematurely, hence. Is there a way to set a default runtime limit for all jobs? I don't want to go through the hassle of modifying all of the job specifications. I tried using --maxJobDuration, but it had no effect.

Many thanks in advance,
Sebastian