Ray.jl

The Ray.jl package provides a Julia language interface for Ray.io workloads.

Make sure the appropriate Python environment (i.e. source venv/bin/activate) is active (wherever you pip installed the Ray CLI) and then do:

ray start --head

to start and

ray stop

to stop.

The directory /tmp/ray/session_latest/logs contains logs for the current or last ran ray backend.

The raylet.err is particularly informative when debugging workers failing to start, since error output before connecting to the Ray server is printed there.

Driver logs generated by Ray are printed in julia-core-driver-$(JOBID)_$(PID).log, and julia worker logs are in julia_worker_$(PID).log (although this may change).

Check the raylet logs in /tmp/ray/session_latest/logs/raylet.err. If you see something about Revise (or another package) not being found, make sure you're not doing something like unconditionally using Revise in your ~/.julia/config/startup.jl file; it's generally a good idea to wrap any usings in your startup.jl in a try/catch block like the Revise docs recommend.