You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Right now the import thread fetches remote function definitions, actor definitions, custom class definitions (needed for custom serialization callbacks). It should be possible to get rid of the extra thread and to simply fetch these values from Redis when they are needed (though that could add latency the first time a remote function is invoked on a worker).
This would also address an issue brought up in #824.
The text was updated successfully, but these errors were encountered:
Thx, @robertnishihara . This actually reminds me another problem - do we have GC capability in python for remote functions? I.e., whether we support long running services for people continuously submitting new ray jobs (we did that on the java side and we want to co-deploy the python as well, as a shared ray job cluster)?
Currently there is no GC for remote functions in Redis (or in the Python worker processes), but it would not too challenging to associate each remote function with the ID of the driver that defined it and to remove it when the driver exits.
Right now the import thread fetches remote function definitions, actor definitions, custom class definitions (needed for custom serialization callbacks). It should be possible to get rid of the extra thread and to simply fetch these values from Redis when they are needed (though that could add latency the first time a remote function is invoked on a worker).
This would also address an issue brought up in #824.
The text was updated successfully, but these errors were encountered: