ray: [runtime_env] Actor tasks aren't failed correctly when missing runtime_env dependencies

Actor tasks aren’t failed correctly when missing runtime_env dependencies:


@ray.remote(runtime_env={"pip": ["this-does-not-exist"]})
a = A.remote()

# Should fail, but hangs!
ray.get(a.f.remote())

About this issue

  • Original URL
  • State: closed
  • Created 3 years ago
  • Comments: 15 (15 by maintainers)

Most upvoted comments

Note that @rkooo567’s PR doesn’t print a RuntimeEnvSetupError for this case, if you were able to do that it would be good to still get it merged (#19824)

Yeah my solution is half way fixing (also it is fixed only for raylet based actor scheduler not the gcs based), so you can keep working on that PR!

@rkooo567 @edoakes Yeah my implementation will print RuntimeEnvSetupError. I can make a follow-up PR.

@rkooo567 Never mind 😃, next time let’s assign an issue before solving it(this issue wasn’t assigned to me in Github when you are working on it, we just talked in slack). When developing let’s search issue before working. Then we won’t do duplicate work.