Open kevin85421 opened 3 weeks ago
cc @jackhumphries
This is the intended behavior. There is one MutableObjectManager instance per actor, and this manager expects that an actor will only read a mutable object once. Thus, the actor is blocked on the next read until a write occurs.
What happened + What you expected to happen
ray.get(actor2.read.remote())
will get stuck inret = ray.get(self._reader_ref)
in the Channel'sbegin_read
function.Versions / Dependencies
nightly
Reproduction script
See the above section.
Issue Severity
None