mirror of
https://github.com/gryf/coach.git
synced 2026-03-06 01:05:47 +01:00
Integrate coach.py params with distributed Coach. (#42)
* Integrate coach.py params with distributed Coach. * Minor improvements - Use enums instead of constants. - Reduce code duplication. - Ask experiment name with timeout.
This commit is contained in:
Balaji Subramaniam
committed by
GitHub
GitHub
parent
95b4fc6888
commit
7e7006305a
@@ -2,6 +2,7 @@ import os
|
||||
import uuid
|
||||
import json
|
||||
import time
|
||||
from enum import Enum
|
||||
from typing import List
|
||||
from configparser import ConfigParser, Error
|
||||
from rl_coach.orchestrators.deploy import Deploy, DeployParameters
|
||||
@@ -12,10 +13,19 @@ from rl_coach.data_stores.data_store import DataStoreParameters
|
||||
from rl_coach.data_stores.data_store_impl import get_data_store
|
||||
|
||||
|
||||
class RunType(Enum):
|
||||
ORCHESTRATOR = "orchestrator"
|
||||
TRAINER = "trainer"
|
||||
ROLLOUT_WORKER = "rollout-worker"
|
||||
|
||||
def __str__(self):
|
||||
return self.value
|
||||
|
||||
|
||||
class RunTypeParameters():
|
||||
|
||||
def __init__(self, image: str, command: list(), arguments: list() = None,
|
||||
run_type: str = "trainer", checkpoint_dir: str = "/checkpoint",
|
||||
run_type: str = str(RunType.TRAINER), checkpoint_dir: str = "/checkpoint",
|
||||
num_replicas: int = 1, orchestration_params: dict=None):
|
||||
self.image = image
|
||||
self.command = command
|
||||
@@ -97,12 +107,12 @@ class Kubernetes(Deploy):
|
||||
|
||||
def deploy_trainer(self) -> bool:
|
||||
|
||||
trainer_params = self.params.run_type_params.get('trainer', None)
|
||||
trainer_params = self.params.run_type_params.get(str(RunType.TRAINER), None)
|
||||
if not trainer_params:
|
||||
return False
|
||||
|
||||
trainer_params.command += ['--memory-backend-params', json.dumps(self.params.memory_backend_parameters.__dict__)]
|
||||
trainer_params.command += ['--data-store-params', json.dumps(self.params.data_store_params.__dict__)]
|
||||
trainer_params.command += ['--memory_backend_params', json.dumps(self.params.memory_backend_parameters.__dict__)]
|
||||
trainer_params.command += ['--data_store_params', json.dumps(self.params.data_store_params.__dict__)]
|
||||
|
||||
name = "{}-{}".format(trainer_params.run_type, uuid.uuid4())
|
||||
|
||||
@@ -175,13 +185,13 @@ class Kubernetes(Deploy):
|
||||
|
||||
def deploy_worker(self):
|
||||
|
||||
worker_params = self.params.run_type_params.get('worker', None)
|
||||
worker_params = self.params.run_type_params.get(str(RunType.ROLLOUT_WORKER), None)
|
||||
if not worker_params:
|
||||
return False
|
||||
|
||||
worker_params.command += ['--memory-backend-params', json.dumps(self.params.memory_backend_parameters.__dict__)]
|
||||
worker_params.command += ['--data-store-params', json.dumps(self.params.data_store_params.__dict__)]
|
||||
worker_params.command += ['--num-workers', '{}'.format(worker_params.num_replicas)]
|
||||
worker_params.command += ['--memory_backend_params', json.dumps(self.params.memory_backend_parameters.__dict__)]
|
||||
worker_params.command += ['--data_store_params', json.dumps(self.params.data_store_params.__dict__)]
|
||||
worker_params.command += ['--num_workers', '{}'.format(worker_params.num_replicas)]
|
||||
|
||||
name = "{}-{}".format(worker_params.run_type, uuid.uuid4())
|
||||
|
||||
@@ -255,7 +265,7 @@ class Kubernetes(Deploy):
|
||||
pass
|
||||
|
||||
def trainer_logs(self):
|
||||
trainer_params = self.params.run_type_params.get('trainer', None)
|
||||
trainer_params = self.params.run_type_params.get(str(RunType.TRAINER), None)
|
||||
if not trainer_params:
|
||||
return
|
||||
|
||||
@@ -313,7 +323,7 @@ class Kubernetes(Deploy):
|
||||
return
|
||||
|
||||
def undeploy(self):
|
||||
trainer_params = self.params.run_type_params.get('trainer', None)
|
||||
trainer_params = self.params.run_type_params.get(str(RunType.TRAINER), None)
|
||||
api_client = k8sclient.AppsV1Api()
|
||||
delete_options = k8sclient.V1DeleteOptions()
|
||||
if trainer_params:
|
||||
@@ -321,7 +331,7 @@ class Kubernetes(Deploy):
|
||||
api_client.delete_namespaced_deployment(trainer_params.orchestration_params['deployment_name'], self.params.namespace, delete_options)
|
||||
except k8sclient.rest.ApiException as e:
|
||||
print("Got exception: %s\n while deleting trainer", e)
|
||||
worker_params = self.params.run_type_params.get('worker', None)
|
||||
worker_params = self.params.run_type_params.get(str(RunType.ROLLOUT_WORKER), None)
|
||||
if worker_params:
|
||||
try:
|
||||
api_client.delete_namespaced_deployment(worker_params.orchestration_params['deployment_name'], self.params.namespace, delete_options)
|
||||
|
||||
@@ -1,120 +0,0 @@
|
||||
import argparse
|
||||
|
||||
from rl_coach.orchestrators.kubernetes_orchestrator import KubernetesParameters, Kubernetes, RunTypeParameters
|
||||
from rl_coach.memories.backend.redis import RedisPubSubMemoryBackendParameters
|
||||
from rl_coach.data_stores.data_store import DataStoreParameters
|
||||
from rl_coach.data_stores.s3_data_store import S3DataStoreParameters
|
||||
from rl_coach.data_stores.nfs_data_store import NFSDataStoreParameters
|
||||
|
||||
|
||||
def main(preset: str, image: str='ajaysudh/testing:coach', num_workers: int=1, nfs_server: str=None, nfs_path: str=None,
|
||||
memory_backend: str=None, data_store: str=None, s3_end_point: str=None, s3_bucket_name: str=None,
|
||||
s3_creds_file: str=None, policy_type: str="OFF"):
|
||||
rollout_command = ['python3', 'rl_coach/rollout_worker.py', '-p', preset, '--policy-type', policy_type]
|
||||
training_command = ['python3', 'rl_coach/training_worker.py', '-p', preset, '--policy-type', policy_type]
|
||||
|
||||
memory_backend_params = None
|
||||
if memory_backend == "redispubsub":
|
||||
memory_backend_params = RedisPubSubMemoryBackendParameters()
|
||||
|
||||
ds_params_instance = None
|
||||
if data_store == "s3":
|
||||
ds_params = DataStoreParameters("s3", "", "")
|
||||
ds_params_instance = S3DataStoreParameters(ds_params=ds_params, end_point=s3_end_point, bucket_name=s3_bucket_name,
|
||||
creds_file=s3_creds_file, checkpoint_dir="/checkpoint")
|
||||
elif data_store == "nfs":
|
||||
ds_params = DataStoreParameters("nfs", "kubernetes", {"namespace": "default"})
|
||||
ds_params_instance = NFSDataStoreParameters(ds_params)
|
||||
|
||||
worker_run_type_params = RunTypeParameters(image, rollout_command, run_type="worker", num_replicas=num_workers)
|
||||
trainer_run_type_params = RunTypeParameters(image, training_command, run_type="trainer")
|
||||
|
||||
orchestration_params = KubernetesParameters([worker_run_type_params, trainer_run_type_params],
|
||||
kubeconfig='~/.kube/config', nfs_server=nfs_server, nfs_path=nfs_path,
|
||||
memory_backend_parameters=memory_backend_params,
|
||||
data_store_params=ds_params_instance)
|
||||
orchestrator = Kubernetes(orchestration_params)
|
||||
if not orchestrator.setup():
|
||||
print("Could not setup")
|
||||
return
|
||||
|
||||
if orchestrator.deploy_trainer():
|
||||
print("Successfully deployed")
|
||||
else:
|
||||
print("Could not deploy")
|
||||
return
|
||||
|
||||
if orchestrator.deploy_worker():
|
||||
print("Successfully deployed")
|
||||
else:
|
||||
print("Could not deploy")
|
||||
return
|
||||
|
||||
try:
|
||||
orchestrator.trainer_logs()
|
||||
except KeyboardInterrupt:
|
||||
pass
|
||||
orchestrator.undeploy()
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument('--image',
|
||||
help="(string) Name of a docker image.",
|
||||
type=str,
|
||||
required=True)
|
||||
parser.add_argument('-p', '--preset',
|
||||
help="(string) Name of a preset to run (class name from the 'presets' directory).",
|
||||
type=str,
|
||||
required=True)
|
||||
parser.add_argument('--memory-backend',
|
||||
help="(string) Memory backend to use.",
|
||||
type=str,
|
||||
choices=['redispubsub'],
|
||||
default="redispubsub")
|
||||
parser.add_argument('--data-store',
|
||||
help="(string) Data store to use.",
|
||||
type=str,
|
||||
choices=['s3', 'nfs'],
|
||||
default="s3")
|
||||
parser.add_argument('--nfs-server',
|
||||
help="(string) Addresss of the nfs server.",
|
||||
type=str,
|
||||
required=False)
|
||||
parser.add_argument('--nfs-path',
|
||||
help="(string) Exported path for the nfs server.",
|
||||
type=str,
|
||||
required=False)
|
||||
parser.add_argument('--s3-end-point',
|
||||
help="(string) S3 endpoint to use when S3 data store is used.",
|
||||
type=str,
|
||||
required=False)
|
||||
parser.add_argument('--s3-bucket-name',
|
||||
help="(string) S3 bucket name to use when S3 data store is used.",
|
||||
type=str,
|
||||
required=False)
|
||||
parser.add_argument('--s3-creds-file',
|
||||
help="(string) S3 credentials file to use when S3 data store is used.",
|
||||
type=str,
|
||||
required=False)
|
||||
parser.add_argument('--num-workers',
|
||||
help="(string) Number of rollout workers.",
|
||||
type=int,
|
||||
required=False,
|
||||
default=1)
|
||||
parser.add_argument('--policy-type',
|
||||
help="(string) The type of policy: OFF/ON.",
|
||||
type=str,
|
||||
choices=['ON', 'OFF'],
|
||||
default='OFF')
|
||||
|
||||
# parser.add_argument('--checkpoint_dir',
|
||||
# help='(string) Path to a folder containing a checkpoint to write the model to.',
|
||||
# type=str,
|
||||
# default='/checkpoint')
|
||||
args = parser.parse_args()
|
||||
|
||||
main(preset=args.preset, image=args.image, nfs_server=args.nfs_server, nfs_path=args.nfs_path,
|
||||
memory_backend=args.memory_backend, data_store=args.data_store, s3_end_point=args.s3_end_point,
|
||||
s3_bucket_name=args.s3_bucket_name, s3_creds_file=args.s3_creds_file, num_workers=args.num_workers,
|
||||
policy_type=args.policy_type)
|
||||
@@ -1,18 +0,0 @@
|
||||
from rl_coach.orchestrators.kubernetes_orchestrator import KubernetesParameters, Kubernetes
|
||||
|
||||
# image = 'gcr.io/constant-cubist-173123/coach:latest'
|
||||
image = 'ajaysudh/testing:coach'
|
||||
command = ['python3', 'rl_coach/rollout_worker.py', '-p', 'CartPole_DQN_distributed']
|
||||
# command = ['sleep', '10h']
|
||||
|
||||
params = KubernetesParameters(image, command, kubeconfig='~/.kube/config', redis_ip='redis-service.ajay.svc', redis_port=6379, num_workers=1)
|
||||
# params = KubernetesParameters(image, command, kubeconfig='~/.kube/config')
|
||||
|
||||
obj = Kubernetes(params)
|
||||
if not obj.setup():
|
||||
print("Could not setup")
|
||||
|
||||
if obj.deploy():
|
||||
print("Successfully deployed")
|
||||
else:
|
||||
print("Could not deploy")
|
||||
Reference in New Issue
Block a user