easyr1/verl/utils/checkpoint/checkpoint_manager.py

# Copyright 2024 Bytedance Ltd. and/or its affiliates
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
import random
import re
import shutil
import tempfile
from abc import ABC, abstractmethod
from typing import Any, Dict, Optional, Union

import numpy as np
import torch
import torch.distributed as dist
from filelock import FileLock
from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from transformers import PreTrainedTokenizer, ProcessorMixin


CHECKPOINT_TRACKER = "latest_global_step.txt"


class BaseCheckpointManager(ABC):
    """
    A checkpoint manager that saves and loads
    - model
    - optimizer
    - lr_scheduler
    - extra_states
    in a SPMD way.

    We save
    - sharded model states and optimizer states
    - full lr_scheduler states
    - huggingface tokenizer and config for ckpt merge
    """

    def __init__(
        self,
        model: FSDP,
        optimizer: torch.optim.Optimizer,
        lr_scheduler: torch.optim.lr_scheduler.LRScheduler,
        processing_class: Union[PreTrainedTokenizer, ProcessorMixin],
    ):
        self.model = model
        self.optimizer = optimizer
        self.lr_scheduler = lr_scheduler
        self.processing_class = processing_class

        assert isinstance(self.model, FSDP)
        self.rank = dist.get_rank()
        self.world_size = dist.get_world_size()

    @abstractmethod
    def load_checkpoint(self, *args, **kwargs):
        raise NotImplementedError

    @abstractmethod
    def save_checkpoint(self, *args, **kwargs):
        raise NotImplementedError

    @staticmethod
    def local_mkdir(path: str) -> str:
        if not os.path.isabs(path):
            working_dir = os.getcwd()
            path = os.path.join(working_dir, path)

        # Using hash value of path as lock file name to avoid long file name
        lock_filename = f"ckpt_{hash(path) & 0xFFFFFFFF:08x}.lock"
        lock_path = os.path.join(tempfile.gettempdir(), lock_filename)

        try:
            with FileLock(lock_path, timeout=60):
                os.makedirs(path, exist_ok=True)
        except Exception as e:
            print(f"Warning: Failed to acquire lock for {path}: {e}")
            os.makedirs(path, exist_ok=True)  # even if the lock is not acquired, try to create the directory

        return path

    @staticmethod
    def get_rng_state() -> Dict[str, Any]:
        rng_state = {
            "cpu": torch.get_rng_state(),
            "cuda": torch.cuda.get_rng_state(),
            "numpy": np.random.get_state(),
            "random": random.getstate(),
        }
        return rng_state

    @staticmethod
    def load_rng_state(rng_state: Dict[str, Any]):
        torch.set_rng_state(rng_state["cpu"])
        torch.cuda.set_rng_state(rng_state["cuda"])
        np.random.set_state(rng_state["numpy"])
        random.setstate(rng_state["random"])


def find_latest_ckpt_path(path: Optional[str] = None, directory_format: str = "global_step_{}") -> Optional[str]:
    if path is None:
        return None

    tracker_file = get_checkpoint_tracker_filename(path)
    if not os.path.exists(tracker_file):
        print("Checkpoint tracker file does not exist: %s", tracker_file)
        return None

    with open(tracker_file, "rb") as f:
        iteration = int(f.read().decode())

    ckpt_path = os.path.join(path, directory_format.format(iteration))
    if not os.path.exists(ckpt_path):
        print("Checkpoint does not exist: %s", ckpt_path)
        return None

    print("Found checkpoint: %s", ckpt_path)
    return ckpt_path


def get_checkpoint_tracker_filename(root_path: str) -> str:
    """
    Tracker file rescords the latest chckpoint during training to restart from.
    """
    return os.path.join(root_path, CHECKPOINT_TRACKER)

import os
import shutil
import re
import time
from watchdog.observers import Observer
from watchdog.events import FileSystemEventHandler

def remove_obsolete_ckpt(
    path: str,
    global_step: int,
    save_limit: int = -1,
    directory_format: str = "global_step_{}",
    protected_steps: set = {46, 23, 69, 92, 115, 138, 161, 184, 230, 276, 322},
    watch_mode: bool = False,
    cleanup_interval: int = 300
):
    """
    Remove the obsolete checkpoints that exceed the save_limit with enhanced features:
    - Protected steps that won't be deleted
    - Watch mode for automatic cleanup
    - Time-based cleanup option
    
    Args:
        path: Directory containing checkpoints
        global_step: Current training step
        save_limit: Maximum number of old checkpoints to keep
        directory_format: Format string for checkpoint directories
        protected_steps: Set of step numbers to never delete
        watch_mode: Enable automatic directory watching
        cleanup_interval: Seconds between cleanups in watch mode
    """
    if save_limit <= 0:
        return

    if not os.path.exists(path):
        return
    steady_nev = os.getenv("steady", "F")
    if steady_nev == "train_and_aime_dapo":
        protected_steps = {50, 100, 150, 200, 250, 25, 75, 125, 175, 225}
    elif "thinkprune" in  steady_nev:
        protected_steps = {59, 118, 177, 236, 354, 432, 540, 648}

    # Define the cleanup function that can be called standalone or by the watcher
    def _cleanup_checkpoints():
        pattern = re.escape(directory_format).replace(r"\{\}", r"(\d+)")
        ckpt_folders = []
        
        # Find all matching checkpoint folders
        for folder in os.listdir(path):
            if match := re.match(pattern, folder):
                step = int(match.group(1))
                if step < global_step:
                    ckpt_folders.append((step, folder))
        
        # Sort checkpoints by step number (newest first)
        ckpt_folders.sort(reverse=True)
        
        # Remove checkpoints beyond save_limit, skipping protected ones
        removed_any = False
        for _, folder in ckpt_folders[save_limit - 1:]:
            folder_path = os.path.join(path, folder)
            # if f"global_step_{int(folder.split('_')[-1])}" not in {f"global_step_{s}" for s in protected_steps}:
            step_num = int(folder.split('_')[-1])
            if step_num % 10 != 0:
                shutil.rmtree(folder_path, ignore_errors=True)
                print(f"Removed obsolete checkpoint: {folder_path}")
                removed_any = True
            else:
                from ...trainer.model_merger import merge_and_save_model, reorganize_folders
                models_path = os.path.join(folder_path, "models")
                if not os.path.exists(models_path):
                    actor_path = os.path.join(folder_path, "actor")
                    merge_and_save_model(actor_path)
                    reorganize_folders(folder_path)

        
        if not removed_any:
            print(f"No checkpoints needed removal (kept {min(save_limit, len(ckpt_folders))}/{len(ckpt_folders)})")

    # If not in watch mode, just do one cleanup
    if not watch_mode:
        _cleanup_checkpoints()
        return

    # Watch mode implementation
    class CheckpointHandler(FileSystemEventHandler):
        # 当文件被创建时调用
        def on_created(self, event):
            # 如果创建的是目录，并且目录名符合指定格式
            if event.is_directory and re.match(
                re.escape(directory_format).replace(r"\{\}", r"\d+"), 
                os.path.basename(event.src_path)
            ):
                # 清理检查点
                _cleanup_checkpoints()

    print(f"Starting checkpoint watcher for {path} (cleanup every {cleanup_interval}s)")
    event_handler = CheckpointHandler()
    observer = Observer()
    observer.schedule(event_handler, path, recursive=False)
    observer.start()

    try:
        while True:
            _cleanup_checkpoints()
            time.sleep(cleanup_interval)
    except KeyboardInterrupt:
        observer.stop()
    observer.join()