Avoid locking database when importing properties in a MPI session

[Martin-Rey]

This suggestion from @apontzen fixes #180.

Is there a good way to test for this parallel-only behaviour? I could find some tests where the PropertyImporter is used in an integrated way (test_ahf_trees, test_stat_files), but it does not seem like there is a dedicated test suite for its internal capabilities.

[apontzen]

It would be useful to have a clearer test suite for PropertyImporter. As for parallel issues, they are notoriously hard to test unfortunately. The underlying locking mechanisms are tested in test_parallel_tasks, which is hard enough. If you can come up with a neat and reliable way to test the integrated behaviour robustness to SQLite locks that would be great, but I don't think it's an easy task.

[Martin-Rey]

Ok, I wouldn't know where to start on parallel testing unfortunately.

As for a more general test suite for the PropertyImporter, I have an unfinished PR upgrading its capabilities to address #162. Maybe it makes more sense to add the test suite there when we are actually developing the class, rather than here?

[apontzen]

Yes, I'm happy with doing it that way

[Martin-Rey]

Ok, then happy to merge once the tests have finished (is it usual for them to take that long now?)

[apontzen]

Yes there is now a full integration test that takes about 1hr to run