search

ECP-VeloC / redset

Low-level distributed erasure coding library to protect datasets of MPI applications
MIT License
2 stars 3 forks source link

reduce number of nodes for tests to 1 node #41

Open adammoody opened 2 years ago

adammoody commented 2 years ago

This one will take a bit more effort, because it uses hostname to define failure groups. This will need to be changed to some string that encodes a rank id or something instead.

gonsie commented 2 years ago

which tests require more than 1 node? everything but single?

It looks like test_reset has 4 tests:

single partner xor rs We could put a simple MPI check before running the multi-node test. an all reduce (maybe comm split?) on the hostname. If we don't have multiple nodes we shouldn't run the tests.

alternatively, we could maybe put a toggle in to say whether we expect the test to pass or fail?

adammoody commented 2 years ago

My hope is that we can rewrite the test program so that all of the tests can run on a single node, but we have some work to do to accomplish that, if it's possible. Until then, we can just keep things simple and assume that all tests require multiple nodes (even if a few don't).