scripts/danbooru/danbooru-check.py

import time, os, json
from tqdm import tqdm
from curl_cffi import requests
import concurrent.futures
from pathlib import Path
import tarfile
import shutil
from huggingface_hub import HfApi

os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"

def main():
    danbooru_path = Path("danbooru")
    selected_range = set(['%03d' % i for i in [400]])
    notexists, total = 0, 0
    notexistsid = {}
    with open("posts.json", "r") as f:
        bar = tqdm(desc="Indexing files", ascii=True, leave=False)
        cache = {}
        data_size = {}
        for d in selected_range:
            data = []
            fp = ('0' + d)
            for file_path in (danbooru_path / fp).iterdir():
                if file_path.is_file():
                    data.append(file_path.stem)
                    data_size[file_path.stem] = file_path.stat().st_size
                    bar.update(1)
            cache[fp] = set(data)

        bar = tqdm(desc="Checking files", total=6_857_737, ascii=True, leave=False)
        for line in f:
            post = json.loads(line)
            file_id = post['id']
            cutoff = str(file_id)[-3:]
            if cutoff not in selected_range:
                bar.update(1)
                continue
            cutoff = '0' + cutoff
            assert (danbooru_path / cutoff).exists(), f"{(danbooru_path / cutoff)} not exixts"
            exists = str(file_id) in cache[cutoff] and (data_size[str(file_id)] == post["file_size"] or int(file_id) < 5_020_995)
            total += 1
            if not exists and "file_url" in post:
                notexists += 1
                if cutoff not in notexistsid:
                    notexistsid[cutoff] = []
                notexistsid[cutoff].append((
                    file_id, cutoff, post["file_url"],
                ))
                # print(post["file_url"])
            bar.update(1)
            bar.set_postfix_str(f"not exists: {notexists}")

    bar.close()
    print(f"not exists: {notexists}, total: {total}")


    with concurrent.futures.ThreadPoolExecutor(max_workers=4) as tar_executor:
        for d in selected_range:
            cut = '0' + d
            if cut not in notexistsid:
                tar_executor.submit(archive_and_upload, Path("danbooru") / cut, cut)

        for key, group in notexistsid.items():
            keybar = tqdm(desc=f"Downloading files in key={key}", total=len(group), position=1, ascii=True, leave=False)
            ok = False
            while not ok:
                with concurrent.futures.ThreadPoolExecutor(max_workers=3) as executor:
                    for file_id, cutoff, file_url in group:
                        executor.submit(download, file_id, cutoff, file_url, keybar)
                ok = True
                for file_id, cutoff, file_url in group:
                    suffix = Path(file_url).suffix
                    if file_url != "" and not Path(f"danbooru/{cutoff}/{file_id}{suffix}").is_file():
                        ok = False
            tar_executor.submit(archive_and_upload, Path(f"danbooru/{cutoff}"), cutoff)
            print(f"Finished download group {cutoff}")
            keybar.close()

def rm_tree(pth: Path):
    for child in pth.iterdir():
        if child.is_file():
            child.unlink()
        else:
            rm_tree(child)
    pth.rmdir()

def archive_and_upload(dirname, name):
    tar_name = Path("danbooru-tars") / f"data-{name}.tar"
    # Check if the directory exists
    if not os.path.isdir(dirname):
        print("The specified directory does not exist.")
        return
    
    # Create the tar file
    print(f"Creating {tar_name}")
    with tarfile.open(tar_name, "w") as tar:
        # Sort and add files to the tar file
        for root, dirs, files in os.walk(dirname):
            # Sort files for consistent ordering
            for file in tqdm(sorted(files), desc=f"Creating {tar_name}", ascii=True):
                full_path = os.path.join(root, file)
                # Add the file to the tar archive
                tar.add(full_path, arcname=file)
    
    # Remove the original directory after archiving
    rm_tree(dirname)
    print(f"The directory {dirname} has been removed.")

    api = HfApi()
    print(api.upload_file(
        path_or_fileobj=tar_name,
        path_in_repo=f"original/data-{name}.tar",
        repo_id="nyanko7/danbooru2023",
        repo_type="dataset",
    ))
    Path(tar_name).unlink()

def download(idx, cutoff, file_url, bar):
    suffix = Path(file_url).suffix
    max_attempts = 3  # specify the maximum number of attempts

    for attempt in range(max_attempts):
        try:
            r = requests.get(file_url, impersonate="chrome110", timeout=120)
            if r.status_code == 200:
                with open(f"danbooru/{cutoff}/{idx}{suffix}", "wb") as f:
                    f.write(r.content)
                break  # if the download is successful, break the loop
            else:
                print(f"Attempt {attempt+1} failed to download {file_url}: error {r.status_code}")
        except Exception as e:
            print(f"Attempt {attempt+1} failed to download {file_url}: error {e}")

        time.sleep(1)  # wait for 1 second before the next attempt
        if attempt+1 == max_attempts:
            print(f"Failed to download {file_url} after {max_attempts} attempts.")
    bar.update(1)

if __name__ == "__main__":
    main()