The object store by default uses up to a certain percentage of the node memory. After that, it spills to disk. Object store related operations will raise an error if the disk is filling up past a certain threshold.
https://docs.ray.io/en/latest/ray-core/objects/object-spilling.html#object-spilling
In short, the object store reduces the available heap memory, and the actor or task is the one that is likely to be OOM-killed