Yi Cheng
Yi Cheng
[Core][deprecate run_on_all_workers 1/n] set worker's sys.path through JobConfig._py_driver_sys_path
Thanks for fixing this. As synced offline, my only concern is that we make the code_search_path and the new introduced ones doing similar works. Maybe we should try to unify...
[Core][deprecate run_on_all_workers 1/n] set worker's sys.path through JobConfig._py_driver_sys_path
Btw, could you also delete import thread which is not useful anymore with this work.
Question: will this slow down the perfs? I think this adds runtime env to all task specs (previously, only one). Do you mind benchmarking the perf regression? Besides this, do...
@scv119 the default microbenchmark doesn't use job config. What concerns me is this: ``` https://sourcegraph.com/github.com/ray-project/ray@master/-/blob/src/ray/protobuf/gcs.proto?L311 ``` Could you add some runtime envs there, for example, a list of OS envs...
``` import ray ray.init() @ray.remote def process(data): return b"\0" * 100_000_000 while True: data = ray.put(b"\0" * 100_000_000) ref = process.remote(data) ray.get(ref) del ref ``` Won't have the leak. Maybe...
``` [2023-01-04 21:17:14,005 D 3290269 3290314] (raylet) store.cc:401: Received create request for object 6d83b3e72b6da8ce4f1fb3b11f2a02d5aeeedd2e6459ec7100000000 assigned request ID 1, 8 bytes [2023-01-04 21:17:14,005 D 3290269 3290314] (raylet) object_lifecycle_manager.cc:38: attempting to create...
``` [2023-01-04 22:03:49,667 D 3319491 3319536] (raylet) store.cc:401: Received create request for object c8ef45ccd0112571ffffffffffffffffffffffff0100000001000000 assigned request ID 19, 100000003 bytes [2023-01-04 22:03:49,667 D 3319491 3319536] (raylet) object_lifecycle_manager.cc:38: attempting to create...
I feel the root cause is related to the lineage reconstruction. Basically, A->B and when B is deleted, A is still there because the lineage count 1 (fail to decrease)....
verified, turn off lineage reconstruction spilling disappeared: ``` RAY_lineage_pinning_enabled=0 python ./leak.py ```
@ericl verified that your theory is correct. Summary: - del ref ==> no leak - del data or del nothing ==> leak - disable lineage ==> no leak I think...