207 lines
6.6 KiB
Python
207 lines
6.6 KiB
Python
import sys
|
|
import time
|
|
from pathlib import Path
|
|
from typing import List
|
|
|
|
import pytest
|
|
|
|
import ray
|
|
import ray.train
|
|
from ray.cluster_utils import Cluster
|
|
from ray.train.tests.util import create_dict_checkpoint, load_dict_checkpoint
|
|
from ray.train.v2._internal.constants import HEALTH_CHECK_INTERVAL_S_ENV_VAR
|
|
from ray.train.v2.api.data_parallel_trainer import DataParallelTrainer
|
|
|
|
|
|
@pytest.fixture
|
|
def cluster():
|
|
cluster = Cluster(initialize_head=True, head_node_args=dict(num_cpus=0))
|
|
cluster.wait_for_nodes()
|
|
ray.init(
|
|
address=cluster.address,
|
|
runtime_env={"working_dir": str(Path(__file__).parent)},
|
|
)
|
|
yield cluster
|
|
ray.shutdown()
|
|
cluster.shutdown()
|
|
|
|
|
|
def train_fn(config: dict):
|
|
train_context = ray.train.get_context()
|
|
rank = train_context.get_world_rank()
|
|
|
|
start_epoch = 1
|
|
checkpoint = ray.train.get_checkpoint()
|
|
min_world_size = None
|
|
max_world_size = None
|
|
if checkpoint:
|
|
checkpoint_data = load_dict_checkpoint(checkpoint)
|
|
start_epoch = checkpoint_data["epoch"] + 1
|
|
min_world_size = checkpoint_data.get("min_world_size")
|
|
max_world_size = checkpoint_data.get("max_world_size")
|
|
if rank == 0:
|
|
print("Restoring from epoch: ", start_epoch)
|
|
|
|
for epoch in range(start_epoch, config.get("num_epochs", 60) + 1):
|
|
world_size = train_context.get_world_size()
|
|
if min_world_size is None:
|
|
min_world_size = world_size
|
|
if max_world_size is None:
|
|
max_world_size = world_size
|
|
min_world_size = min(min_world_size, world_size)
|
|
max_world_size = max(max_world_size, world_size)
|
|
# TODO: This test injects errors by "killing nodes," which ungracefully
|
|
# kills processes. This means that any backlog in the checkpoint queue
|
|
# will not be flushed to the controller.
|
|
# This means that the checkpoint populated on restore may not be
|
|
# the most recent one.
|
|
# Set the poll interval < health check interval to reduce the
|
|
# backlog size to mitigate the issue.
|
|
time.sleep(2 * config.get("health_check_interval_s", 1))
|
|
|
|
with create_dict_checkpoint(
|
|
{
|
|
"epoch": epoch,
|
|
"min_world_size": min_world_size,
|
|
"max_world_size": max_world_size,
|
|
}
|
|
) as checkpoint:
|
|
ray.train.report(
|
|
{
|
|
"epoch": epoch,
|
|
"world_size": world_size,
|
|
"min_world_size": min_world_size,
|
|
"max_world_size": max_world_size,
|
|
},
|
|
checkpoint=checkpoint if rank == 0 else None,
|
|
checkpoint_dir_name=f"checkpoint-epoch={epoch}",
|
|
)
|
|
if rank == 0:
|
|
print("Finished epoch: ", epoch)
|
|
|
|
|
|
def test_elastic_training(monkeypatch, tmp_path, cluster):
|
|
"""End to end test for elastic training.
|
|
|
|
This test covers:
|
|
* Elastic startup (0 resources -> min resources)
|
|
* Elastic scale up while running (min resources -> max resources)
|
|
* Elastic scale down due to failure while running
|
|
* Checkpointing + restoration
|
|
* Preemption failure handling
|
|
"""
|
|
unit_time_s = 0.1
|
|
health_check_interval_s = unit_time_s
|
|
elastic_resize_monitor_interval_s = unit_time_s * 10
|
|
num_epochs = 30
|
|
|
|
monkeypatch.setenv(HEALTH_CHECK_INTERVAL_S_ENV_VAR, str(health_check_interval_s))
|
|
|
|
@ray.remote(num_cpus=0)
|
|
def run_training():
|
|
trainer = DataParallelTrainer(
|
|
train_fn,
|
|
train_loop_config={
|
|
"num_epochs": num_epochs,
|
|
"health_check_interval_s": health_check_interval_s,
|
|
},
|
|
scaling_config=ray.train.ScalingConfig(
|
|
num_workers=(4, 32),
|
|
use_gpu=True,
|
|
elastic_resize_monitor_interval_s=elastic_resize_monitor_interval_s,
|
|
),
|
|
run_config=ray.train.RunConfig(
|
|
storage_path=str(tmp_path),
|
|
checkpoint_config=ray.train.CheckpointConfig(num_to_keep=2),
|
|
# NOTE: The outer test script will inject 2 node failures.
|
|
failure_config=ray.train.FailureConfig(max_failures=2),
|
|
),
|
|
)
|
|
return trainer.fit()
|
|
|
|
run_training_future = run_training.remote()
|
|
|
|
start = time.time()
|
|
ALL_NODES = []
|
|
|
|
def print_status(message):
|
|
elapsed = time.time() - start
|
|
print()
|
|
print("-" * 80)
|
|
cluster_resources = {
|
|
resource: value
|
|
for resource, value in ray.cluster_resources().items()
|
|
if resource in ("CPU", "GPU")
|
|
}
|
|
print(f"[elapsed={elapsed:.1f}s] {cluster_resources=}")
|
|
print(message)
|
|
print("-" * 80)
|
|
print()
|
|
|
|
def sleep(num_units):
|
|
time.sleep(unit_time_s * num_units)
|
|
|
|
def add_nodes(gpus: List[int]) -> List:
|
|
added_nodes = []
|
|
for num_gpus in gpus:
|
|
node = cluster.add_node(num_gpus=num_gpus, wait=False)
|
|
added_nodes.append(node)
|
|
|
|
print_status(f"Added {len(gpus)} node(s) with num_gpus: {gpus}")
|
|
cluster.wait_for_nodes()
|
|
return added_nodes
|
|
|
|
def remove_nodes(nodes: List):
|
|
for node in nodes:
|
|
cluster.remove_node(node)
|
|
|
|
cluster.wait_for_nodes()
|
|
print_status(f"Removed nodes: {nodes}")
|
|
|
|
# Wait a bit before adding resources.
|
|
print_status("Waiting for training to start...")
|
|
sleep(8)
|
|
|
|
# Add a node with 4 GPUs
|
|
ALL_NODES.extend(add_nodes([4]))
|
|
|
|
# Wait a bit before adding more resources.
|
|
sleep(8)
|
|
print("Adding 4 GPU node.")
|
|
ALL_NODES.extend(add_nodes([4]))
|
|
sleep(1)
|
|
ALL_NODES.extend(add_nodes([4]))
|
|
# Should not upscale here due to the elastic resize monitor interval.
|
|
|
|
# Should upscale to 12 during this sleep.
|
|
sleep(20)
|
|
|
|
# Kill a node.
|
|
remove_nodes([ALL_NODES.pop(0)])
|
|
sleep(12)
|
|
|
|
# Kill all worker nodes.
|
|
remove_nodes(ALL_NODES)
|
|
ALL_NODES = []
|
|
|
|
sleep(8)
|
|
ALL_NODES.extend(add_nodes(gpus=[1] * 16))
|
|
|
|
sleep(12)
|
|
# 4 extra GPUs shouldn't be used.
|
|
ALL_NODES.extend(add_nodes(gpus=[4] * 4 + [1] * 4))
|
|
|
|
result: ray.train.Result = ray.get(run_training_future)
|
|
|
|
print_status(f"Training finished with result: {result}")
|
|
assert not result.error
|
|
assert result.metrics["min_world_size"] >= 4
|
|
assert result.metrics["max_world_size"] <= 32
|
|
assert result.metrics["max_world_size"] >= result.metrics["min_world_size"]
|
|
assert result.checkpoint
|
|
assert Path(result.checkpoint.path).name == f"checkpoint-epoch={num_epochs}"
|
|
|
|
|
|
if __name__ == "__main__":
|
|
sys.exit(pytest.main(["-v", "-x", __file__]))
|