I have the following production use case and am trying to see if there are other solutions than using ray client as it seems to be limited (for e.g. doesn’t support generators) and the documentation seems to be cautions about using it
I have a cluster of ray workers and a head node
My client process receives requests via grpc
Around every 10min - 1 hour it get a request that in turn will cause a bunch of actors to be created on each of the ray workers
and then around every 1 sec i get a request that in turn I send a remote function call to each of the actors.
Now its a little more complicated than that. But even just with that i don’t see how i can use ray jobs.
I thought maybe i could run my process (that receives the grpc commands) within the ray head node but not sure it is the best practice and how to go about timing it (need to understand that everything is ready before calling ray.init() )
Would appreciate any suggestions or to understand if staying with ray client is my best option in the long run?