I am using Ray Serve to host/serve an NLP based model. My server has 64GB RAM. I start the Ray server using the command:
ray start --head --object-store-memory=2000000000 --port=6379 --dashboard-host=192.168.1.111 --dashboard-port=5002
I get the following error message in my log and my program crashes:
ray.exceptions.ObjectStoreFullError: Failed to put object eff3b67b54ffdcc0615c4dfd5097b9f202000000 in object store because it is full. Object size is 1223738603 bytes. The local object store is full of objects that are still in scope and cannot be evicted. Tip: Use the
ray memorycommand to list active objects in the cluster.
Ray is allocating only 1GB for object storage and the command line argument --object-store-memory=2000000000 does not appear to have any effect. What is the best way to allocate a larger object storage? Also, I see ray support spilling the object to disk but I could not find a way to pass that parameter to the commanline startup command.
Any help is greatly appreciated.