How to get a pull request merged?

RocketRider · May 16, 2023, 4:46pm

What is the best way to get a pull request merged?

I made a nice and easy fix in a pull request, but no reaction from the team.

github.com/ray-project/ray

[RLlib] Fix IMPALA/APPO when using multi GPU setup and Multi-Agent Env

ray-project:master ← RocketRider:patch-1

opened 05:25AM - 08 May 23 UTC

RocketRider

+3 -1

When I tried to run a APPO Training with more than one GPU and a Multi-Agent Env… I got the following error: ``` Failure # 1 (occurred at 2023-05-08_06-54-37) ^[[36mray::CustomTFTrainerSearch.train()^[[39m (pid=1797305, ip=10.10.10.102, repr=CustomTFTrainerSearch) File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/tune/trainable/trainable.py", line 388, in train raise skipped from exception_cause(skipped) File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/tune/trainable/trainable.py", line 385, in train result = self.step() File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/algorithms/algorithm.py", line 801, in step results, train_iter_ctx = self._run_one_training_iteration() File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/algorithms/algorithm.py", line 2842, in _run_one_training_iteration results = self.training_step() File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/algorithms/appo/appo.py", line 372, in training_step train_results = super().training_step() File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/algorithms/impala/impala.py", line 738, in training_step self.update_workers_if_necessary( File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/algorithms/impala/impala.py", line 1165, in update_workers_if_necessary "num_grad_updates_per_policy": { File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/algorithms/impala/impala.py", line 1166, in <dictcomp> pid: local_worker.policy_map[pid].num_grad_updates File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/utils/threading.py", line 24, in wrapper return func(self, *a, **k) File "/home/ubuntu-vm/MyProject/MyProjectim/venv/lib/python3.9/site-packages/ray/rllib/policy/policy_map.py", line 104, in __getitem__ raise KeyError( KeyError: "PolicyID 'default_policy' not found in this PolicyMap! IDs stored in this map: {'BTL'}." ``` "BTL" is the policy_id but at the requested change it was overwrite by the "default_policy". The Code change is tested and now it is running as expected. The default param policy_id= "default_policy" is hiding this issue. For Multi-Agent it is necessary to use the correct policy_id. ## Checks - [ x ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - Testing Strategy - Tested Tune training on 2x 3090 GPU with Multi-Agent Env

Jules_Damji · May 16, 2023, 6:03pm

@RocketRider
cc @arturn @sven1977 Anyone looking into to his PR?

arturn · May 16, 2023, 6:06pm

@RocketRider ,

Thanks for the PR! Sorry for missing it. I’ll look into it today!