alienv fails when launched simultaneously from several terminals

[Barthelemy]

In Terminator I can send commands to all terminals at once. If I do "alienv load QualityControl/latest" it will succeed in one but fail in the others with errors such as :

cp: cannot create regular file ‘/afs/cern.ch/user/b/bvonhall/dev/alice/sw/MODULES/slc7_x86-64/Vc/1.3.2_O2_DAQ-1’: No such file or directory

cp: cannot create regular file ‘/afs/cern.ch/user/b/bvonhall/dev/alice/sw/MODULES/slc7_x86-64/Vc/1.3.2_O2_DAQ-1’: File exists

[ktf]

I think this is due to the fact that we need to copy modulefiles in a specific place in order to be picked up by module. If you do that in parallel, it's probably exposing a race condition. @dberzano what do you think? A possible solution could be to do a first alienv load in a single terminal, and then redo all the others in parallel. Can you check if that works?

[Barthelemy]

@ktf even if I load first in a single terminal I get these errors.

[dberzano]

So in the release validation we actually work it around by using lockfiles. One can use --no-refresh to avoid this, but the problem is that you are not sure you're getting the latest modulefiles. IMHO this should be fixed upstream by us.

[Barthelemy]

ok thank you for assessing the situation.

[olantwin]

We've also recently discovered this issue when using alienv with the CERN HTCondor batch system and the sw directory on a shared file-system.

Until the underlying problem can be fixed, a workaround that works well for us is to evaluate the environment once (e.g. alienv printenv [Package] >> config.sh) and then source this in all processes which need the environment.

I would be happy to help fix the underlying issue, but it might take a while to familiarise myself with the internals of alienv...