Improving startup speed


I’m considering porting my model to rllib, but I’m having quite a long startup time when running ray.init() (about 2-5 seconds).

So my question is simple (I hope). Is it possible to have ray running when shutting down the “development process” so that its faster to prototype and integrate models into the ray/rllib ecosystem?

Could not find any guides on this, but any help would be appreciated :slight_smile:

Hey @perara ,
yes, you could already start the ray cluster before running your script. Like on the command line:

$ ray start --head

Then in your python script, do:

>> ray.init(address="auto")
>> ^^ this should be very fast now ^^