TypeError: cannot pickle '_thread.RLock' object

Ray Libraries (Data, Train, Tune, Serve)
1

I’m trying to use an RLock in an actor. I’m getting

/home/albert/git/reclone/THA-Oracle/oracle/tha_oracle/oracle_face/external_app.py:61: in <module>
    class OracleFace:
/home/albert/miniconda3/envs/tha_py/lib/python3.10/site-packages/ray/serve/api.py:287: in decorator
    frozen_app = cloudpickle.loads(cloudpickle.dumps(app))
/home/albert/miniconda3/envs/tha_py/lib/python3.10/site-packages/ray/cloudpickle/cloudpickle_fast.py:73: in dumps
    cp.dump(obj)
/home/albert/miniconda3/envs/tha_py/lib/python3.10/site-packages/ray/cloudpickle/cloudpickle_fast.py:620: in dump
    return Pickler.dump(self, obj)
E   TypeError: cannot pickle '_thread.RLock' object

Why does it need to pickle the RLock? It’s something used only by the actor, and never passed around.

2

Ah, this might actually be because I’m using an RLock in a decorator of a serve end point. I use a LRU cache on a serve endpoint, however that might not actually be doing what I expect. And I could see how the function decorator itself is serialised.