TimelyDataflow/timely-dataflow

Multi process tests

Opened this issue · 1 comments

There are no tests to make sure timely works correctly for n > 1 cases and this may be an issue sometimes.
For example @utaal recently modified a bit of communication in a fork, and cargo test was ok, but when I tried to run some programs using that forked version of timely, I encountered different problems (crashes & hangs).
This can be avoided by including some tests that verify timely works correctly for n > 1.

utaal commented

To keep this cheap, we may just want a smoke test that starts a small program on 2 processes with 2 workers each. It does require spawning processes which may be a bit tricky to do correctly cross-platform from tests.