Datasets:

ArXiv:
License:
Nymeria / nymeria /download_utils.py
ariakang's picture
updated accompanying tools
a199a9b
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
import hashlib
import json
import shutil
import tempfile
from dataclasses import dataclass
from enum import Enum
from pathlib import Path
from zipfile import is_zipfile, ZipFile
import requests
from loguru import logger
from requests.adapters import HTTPAdapter
from requests.packages.urllib3.util.retry import Retry
from tqdm import tqdm
from .definitions import DataGroups, NYMERIA_VERSION
class DlConfig(Enum):
CHUCK_SIZE_BYTE = 8192
READ_BYTE = 4096
RETRY = 5
BACKOFF_FACTOR = 3
class DlStatus(Enum):
UNKNOWN = None
SUCCESS = "success"
IGNORED = "ignored, file already downloaded"
WARN_NOTFOUND = "warning, missing download link"
ERR_SHA1SUM = "error, sha1sum mismatch"
ERR_MEMORY = "error, insufficient disk space"
ERR_NETWORK = "error, network"
@dataclass
class DlLink:
filename: str
sha1sum: str
file_size_bytes: int
download_url: str
data_group: DataGroups
status: DlStatus = DlStatus.UNKNOWN
def __post_init__(self) -> None:
prefix = f"Nymeria_{NYMERIA_VERSION}_"
if prefix not in self.filename:
self.status = (
f"Version mismatch with the release {NYMERIA_VERSION}. "
f"Please download the latest url json"
)
raise ValueError(self.status)
self.filename = self.filename.replace(prefix, "")
@property
def seq_name(self) -> str:
return "_".join(self.filename.split("_")[0:6])
@property
def logdir(self) -> str:
return "logs"
def __check_outdir(self, outdir: Path) -> None:
assert (
outdir.name == self.seq_name
), f"Output directory name ({outdir.name}) mismatch with sequence {self.seq_name}"
outdir.mkdir(exist_ok=True)
def get(self, outdir: Path, ignore_existing: bool = True) -> None:
"""This function throws error if not successful"""
flag = outdir / self.logdir / self.data_group.name
if flag.is_file() and ignore_existing:
self.status = DlStatus.IGNORED
return
self.__check_outdir(outdir)
with tempfile.TemporaryDirectory() as tmpdir:
tmp_filename = Path(tmpdir) / self.filename
logger.info(f"Download {self.filename} -> {tmp_filename}")
session = requests.Session()
"""
Retry will be triggered for the following cases
(429) Too Many Requests
(500) Internal Server Error
(502) Bad Gateway
(503) Service Unavailable
(504) Gateway Timeout
"""
retries = Retry(
total=DlConfig.RETRY.value,
backoff_factor=DlConfig.BACKOFF_FACTOR.value,
status_forcelist=[429, 500, 502, 503, 504],
)
session.mount("https://", HTTPAdapter(max_retries=retries))
with session.get(self.download_url, stream=True) as r:
free_outdir = shutil.disk_usage(outdir).free
free_tmpdir = shutil.disk_usage(tmpdir).free
if (
free_outdir < self.file_size_bytes
or free_tmpdir < self.file_size_bytes
):
self.status = DlStatus.ERR_MEMORY
raise RuntimeError(
"Insufficient disk space. "
f"Require {self.file_size_bytes}B, "
f"tmpdir availabel {free_tmpdir}B, outdir available {free_outdir}B"
)
with open(tmp_filename, "wb") as f:
sha1 = hashlib.sha1()
progress_bar = tqdm(
total=self.file_size_bytes, unit="iB", unit_scale=True
)
for chunk in r.iter_content(
chunk_size=DlConfig.CHUCK_SIZE_BYTE.value
):
progress_bar.update(len(chunk))
f.write(chunk)
sha1.update(chunk)
computed = sha1.hexdigest()
if self.sha1sum != computed:
self.status = DlStatus.ERR_SHA1SUM
raise RuntimeError(
f"sha1sum mismatch, computed {computed}, expected {self.sha1sum}"
)
progress_bar.close()
try:
r.raise_for_status()
except Exception as e:
self.status = DlStatus.ERR_NETWORK
raise RuntimeError(e)
# move from tmp -> dst
if is_zipfile(tmp_filename):
logger.info("unzip")
with ZipFile(tmp_filename) as zf:
zf.extractall(outdir)
else:
dst_file = outdir / self.data_group.value
dst_file.parent.mkdir(exist_ok=True, parents=True)
shutil.move(src=tmp_filename, dst=dst_file)
logger.info(f"Download {self.filename} -> {outdir}")
self.status = DlStatus.SUCCESS
# create a flag
flag.parent.mkdir(exist_ok=True)
flag.touch()
class DownloadManager:
def __init__(self, url_json: Path, out_rootdir: Path) -> None:
self.url_json = url_json
assert self.url_json.is_file(), f"{self.url_json} not found"
self.out_rootdir = out_rootdir
self.out_rootdir.mkdir(exist_ok=True)
with open(self.url_json, "r") as f:
data = json.load(f)
self._sequences = data.get("sequences", {})
assert len(
self._sequences
), "No sequence found. Please check the json file is correct."
self.__get_data_summary()
self._logs = {}
@property
def sequences(self) -> dict[str, any]:
return self._sequences
@property
def logfile(self) -> Path:
return self.out_rootdir / "download_summary.json"
def __get_data_summary(self):
missing = {x.name: {"count": 0, "sequences": []} for x in DataGroups}
for seq, dgs in self.sequences.items():
for dg in DataGroups:
if dg.name not in dgs:
missing[dg.name]["count"] += 1
missing[dg.name]["sequences"].append(seq)
fname = self.logfile.with_name("data_summary.json")
with open(fname, "w") as f:
json.dump(
{
"missing_files": missing,
"available_sequences": list(self.sequences.keys()),
},
f,
indent=2,
)
logger.info(f"save data summary to {fname}")
def __prepare(
self,
match_key: str,
selected_groups: list["DataGroups"],
) -> set["DataGroups"]:
selected_groups += [DataGroups.LICENSE, DataGroups.metadata_json]
selected_groups = set(selected_groups)
num_seqs = 0
total_gb = 0
self._logs = {}
for seq, dgs in self.sequences.items():
if match_key not in seq:
continue
num_seqs += 1
self._logs[seq] = {}
for dg in selected_groups:
if dg.name not in dgs:
self._logs[seq][dg.name] = DlStatus.WARN_NOTFOUND.value
else:
self._logs[seq][dg.name] = None
dl = DlLink(**{**dgs.get(dg.name, {}), "data_group": dg})
total_gb += dl.file_size_bytes / (2**30)
# populate confirmation msg
msg = "\t" + "\n\t".join([x.value for x in selected_groups])
free_disk_gb = shutil.disk_usage(self.out_rootdir).free / (2**30)
confirm = (
input(
f"Download summary\n"
f" Output rootdir: {self.out_rootdir}\n"
f" Number sequences: {num_seqs}\n"
f" Total memory (GB): {total_gb}\n"
f" Available free disk space (GB): {free_disk_gb}\n"
f" Selected data groups:\n{msg}\n"
f"Proceed: [y/n] "
).lower()
== "y"
)
if not confirm:
exit(1)
return selected_groups
def __logging(self, **kwargs) -> None:
self._logs.update(**kwargs)
with open(self.logfile, "w") as f:
json.dump(self._logs, f, indent=2)
def download(
self,
match_key: str,
selected_groups: list["DataGroups"],
ignore_existing: bool = True,
) -> None:
selected_groups = self.__prepare(match_key, selected_groups)
summary = {x.name: 0 for x in DlStatus}
for seq_name, dgs in self.sequences.items():
if match_key not in seq_name:
continue
outdir = self.out_rootdir / seq_name
for dg in selected_groups:
if dg.name not in dgs:
continue
dl = DlLink(**{**dgs[dg.name], "data_group": dg})
try:
dl.get(outdir, ignore_existing=ignore_existing)
except Exception as e:
logger.error(f"downloading failure:, {e}")
summary[dl.status.name] += 1
self._logs[dl.seq_name][dl.data_group.name] = dl.status.value
self.__logging()
self.__logging(download_summary=summary)
logger.info(f"Dataset download to {self.out_rootdir}")
logger.info(f"Brief download summary: {json.dumps(summary, indent=2)}")
logger.info(f"Detailed summary saved to {self.logfile}")