Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Bypass out-of-sync Gym registry in SubprocVecEnv by resolving EnvSpec #160

Merged
merged 10 commits into from
Jan 22, 2020
Merged

Bypass out-of-sync Gym registry in SubprocVecEnv by resolving EnvSpec #160

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
a1b9c03
make_vec_env: Resolve EnvSpec in main process
shwang Jan 18, 2020
0576bc5
test_scripts: Test Ray+custom_ant+SubprocVecEnv
shwang Jan 18, 2020
f527f0f
make_vec_env: Apply spec.max_episode_steps
shwang Jan 18, 2020
a442b00
test_scripts: Auto-generate ant rollouts
shwang Jan 21, 2020
560c068
make_vec_env: Don't attempt to explain bug
shwang Jan 21, 2020
3f8d29f
test_scripts: Rename helper function
shwang Jan 21, 2020
e7fc11d
Update src/imitation/util/util.py
shwang Jan 22, 2020
365c177
make_vec_env: Expand docstring
shwang Jan 22, 2020
e8061a6
Merge branch 'master' into fix_ray_subproc_gym_registry
shwang Jan 22, 2020
902fe96
make_vec_env: Fix docstring
shwang Jan 22, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 15 additions & 1 deletion src/imitation/util/util.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,15 +49,29 @@ def make_vec_env(env_name: str,
max_episode_steps: If specified, wraps VecEnv in TimeLimit wrapper with
this episode length before returning.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
this episode length before returning.
this episode length before returning. Otherwise, defaults to `max_episode_steps` for `env_name` in the Gym registry.

Copy link
Member Author

@shwang shwang Jan 22, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I expanded this comment a bit more in 902fe96. Wanted to note that the gym registry total timesteps thing is default behavior for gym.make.

"""
# Resolve the spec outside of the subprocess first, so that it is available to
# subprocesses via automatic pickling.
spec = gym.spec(env_name)

def make_env(i, this_seed):
env = gym.make(env_name)
# Previously, we directly called `gym.make(env_name)`.
#
# That direct approach was problematic (especially in combination with Ray)
# because the forkserver from which subprocesses are forked might have been
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are we sure this is what's going on? It's fine as a hypothesis but I don't want to immortalize in a comment something we're not sure about.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm 80% confident the forkserver has a very minimal state, since:

  1. The Python multiprocessing docs say "No unnecessary resources are inherited".
  2. The code seems to use spawn to start the forkserver, which starts a new Python process.
  3. This blog states " Note that children retain a copy of the forkserver state. This state is intended to be relatively simple, but it is possible to adjust this through the multiprocess API through the set_forkserver_preload() method."

With that said, there does seem to be some logic to preload modules (by default, I think, the __main__ module i.e. the entrypoint to the script). So IIUC, forkserver is intended to execute the code of the parent process (i.e. all imports), but should not execute an if __name__ == '__main__'-guarded block (Python docs explicitly state need to have this guard).

Seems plausible that starting forkserver in a Ray worker messes with the autodetection of what to import.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we can't get to the bottom of this, may be better to leave the comment vague and cite this PR.

Copy link
Member Author

@shwang shwang Jan 21, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agree that we should be vague about the cause in the comments and note the PR, I've made these changes.

# spawned before `env_name` was registered in the main process,
# causing `env_name` to never exist in the Gym registry of the forked
# subprocess that is running `make_env(env_name)`.
env = spec.make()

# Seed each environment with a different, non-sequential seed for diversity
# (even if caller is passing us sequentially-assigned base seeds). int() is
# necessary to work around gym bug where it chokes on numpy int64s.
env.seed(int(this_seed))

if max_episode_steps is not None:
env = TimeLimit(env, max_episode_steps)
elif (spec.max_episode_steps is not None) and not spec.tags.get('vnc'):
shwang marked this conversation as resolved.
Show resolved Hide resolved
env = TimeLimit(env, max_episode_steps=spec.max_episode_steps)

# Use Monitor to record statistics needed for Baselines algorithms logging
# Optionally, save to disk
Expand Down
13 changes: 11 additions & 2 deletions tests/test_scripts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,9 +144,9 @@ def test_transfer_learning(tmpdir):
dict(
sacred_ex_name="expert_demos",
base_named_configs=["cartpole", "fast"],
n_seeds=2,
search_space={
"config_updates": {
"seed": tune.grid_search([0, 1]),
"init_rl_kwargs": {
"learning_rate": tune.grid_search([3e-4, 1e-4]),
},
Expand All @@ -169,9 +169,18 @@ def test_transfer_learning(tmpdir):
},
}},
),
# Test that custom environments are passed to SubprocVecEnv in Ray workers.
dict(
sacred_ex_name="train_adversarial",
base_named_configs=["custom_ant", "fast"],
base_config_updates=dict(
init_trainer_kwargs=dict(
parallel=True,
num_vec=2,
)),
),
]


PARALLEL_CONFIG_LOW_RESOURCE = {
# CI server only has 2 cores.
"init_kwargs": {"num_cpus": 2},
Expand Down