Kai-Hsun Chen

Results 323 comments of Kai-Hsun Chen
trafficstars

@stephanie-wang I took a look at the issue. The problem is that any normal `ray.get` call to the actors belonging to the graph will not work because the execution loop...

Maybe it is not worth supporting access to thread-local state if it blocks normal tasks.

https://github.com/ray-project/ray/pull/50032 attempts to move the execution loop to the same thread as the default thread (the thread used when `concurrency_group` is not specified in `options`). * The execution loop can't...

https://github.com/ray-project/ray/pull/50032/commits/b0006205b664cafbeb28e2e1dec3f5f54b18e6d3 this commit moves the constructor to run on the default executor. I ran the following script and the output is attached below. Both `__init__` and `compute` are running on...

The destructor of MyObject will be called immediately after f1 finishes, so f2 cannot access self.thread_data.value. Based on my current observation, our Python thread behaves differently from a normal Python...

I have a reproduction script here: https://gist.github.com/kevin85421/a7f14ea38d64420b105fbd79fd31fb8a ```python with InputNode() as inp: # fail: raise an exception, sleep: sleep $inp seconds dag = MultiOutputNode([b.fail.bind(inp), a.sleep.bind(inp)]) compiled_dag = dag.experimental_compile() ref =...

> Does it work when you do `dag = MultiOutputNode([a.sleep.bind(inp), b.fail.bind(inp)])`? No > I think what we need is try batch waiting with short timeout (and keep checking until every...

I used the same example as https://github.com/ray-project/ray/issues/46284#issuecomment-2195714993. I added a print function, `print("check_status: ", message, status.ok(), status.IsChannelError())`, to `check_status`. https://github.com/ray-project/ray/blob/755a49bbc2345650ff90d4706d9c69f87b8a77bd/python/ray/_raylet.pyx#L560-L561 I have two questions: * I can see `(Actor pid=2054169)...

> Should the link just go directly there? Yes, adding a link directly is fine. KubeRay tries to maintain a single source of truth for documents. We are trying to...