yfinance.download is not thread-safe. concurrent calls with same ticker and different parameters may overwrite results

[danaiger]

Describe bug

When using yfinance.download concurrently (e.g., via concurrent.futures.ThreadPoolExecutor) to download same ticker with different date ranges, results may be silently overwritten.

This is because yfinance.download() internally uses a shared global dictionary, which is not thread-local or protected from concurrent writes.

Here is a simplified version of the internal function used inside download() (from source):

def _download_one(ticker, ...):
    ...
    data = Ticker(ticker).history(...)
    shared._DFS[ticker.upper()] = data
    return data

Since _DFS is a global dictionary, if two threads are downloading the same ticker with different parameters (e.g., different start/end), they may overwrite each other’s results inside _DFS before the final result is returned.

This leads to incorrect results when the user calls yf.download() concurrently.

Notes on Debug Mode

This issue can be harder to reproduce when debug_mode=True, because yfinance.download() automatically sets threads=False internally.
However, even in debug mode, the user’s code may still execute in a multi-threaded context, so this behavior is likely just coincidentally avoided due to timing.
The root cause remains: a global shared dictionary is accessed and written to concurrently. This is a race condition waiting to happen.

Simple code that reproduces your problem

import asyncio
import yfinance as yf
from functools import partial
from concurrent.futures import ThreadPoolExecutor


def download(ticker: str, start: str):
    return yf.download(ticker, start=start, end="2023-01-10", progress=False)


async def main():
    with ThreadPoolExecutor(max_workers=2) as executor:
        loop = asyncio.get_event_loop()
        tasks = [
            loop.run_in_executor(executor, partial(
                download, "AAPL", "2023-01-01")),
            loop.run_in_executor(executor, partial(
                download, "AAPL", "2022-12-01")),
        ]
        results = await asyncio.gather(*tasks)
    for i, df in enumerate(results):
        print(f"\n📊 Result {i + 1}:")
        if df is None or df.empty:
            print("❌ DataFrame is empty!")
        else:
            print(f"  From: {df.index.min().date()} To: {df.index.max().date()}")
            print(f"  First Close: {float(df['Close'].iloc[0].item()):.2f}")
            print(f"  Last Close:  {float(df['Close'].iloc[-1].item()):.2f}")
            print(f"  Rows: {len(df)}")

asyncio.run(main())

Debug log from yf.enable_debug_mode()

DEBUG    Entering download()
DEBUG    Entering download()
YF.download() has changed argument auto_adjust default to True
DEBUG     Disabling multithreading because DEBUG logging enabled
DEBUG     Disabling multithreading because DEBUG logging enabled
DEBUG     Using User-Agent: Mozilla/5.0 (X11; Linux i686; rv:135.0) Gecko/20100101 Firefox/135.0
DEBUG     Entering history()
DEBUG     Entering history()
DEBUG      Entering history()
DEBUG      Entering history()
DEBUG       AAPL: Yahoo GET parameters: {'period1': '2023-01-01 00:00:00-05:00', 'period2': '2023-01-10 00:00:00-05:00', 'interval': '1d', 'includePrePost': False, 'events': 'div,splits,capitalGains'}
DEBUG       Entering get()
DEBUG       AAPL: Yahoo GET parameters: {'period1': '2022-12-01 00:00:00-05:00', 'period2': '2023-01-10 00:00:00-05:00', 'interval': '1d', 'includePrePost': False, 'events': 'div,splits,capitalGains'}
DEBUG        Entering _make_request()
DEBUG       Entering get()
DEBUG         url=https://query2.finance.yahoo.com/v8/finance/chart/AAPL
DEBUG        Entering _make_request()
DEBUG         params=frozendict.frozendict({'period1': 1672549200, 'period2': 1673326800, 'interval': '1d', 'includePrePost': False, 'events': 'div,splits,capitalGains'})
DEBUG         url=https://query2.finance.yahoo.com/v8/finance/chart/AAPL
DEBUG         Entering _get_cookie_and_crumb()
DEBUG         params=frozendict.frozendict({'period1': 1669870800, 'period2': 1673326800, 'interval': '1d', 'includePrePost': False, 'events': 'div,splits,capitalGains'})
DEBUG          cookie_mode = 'basic'
DEBUG         Entering _get_cookie_and_crumb()
DEBUG          Entering _get_cookie_and_crumb_basic()
DEBUG          cookie_mode = 'basic'
DEBUG           crumb = 'mfIDcZ8MHpF'
DEBUG          Exiting _get_cookie_and_crumb_basic()
DEBUG         Exiting _get_cookie_and_crumb()
DEBUG          Entering _get_cookie_and_crumb_basic()
DEBUG           reusing crumb
DEBUG          Exiting _get_cookie_and_crumb_basic()
DEBUG         Exiting _get_cookie_and_crumb()
DEBUG         response code=200
DEBUG        Exiting _make_request()
DEBUG       Exiting get()
DEBUG       AAPL: yfinance received OHLC data: 2023-01-03 14:30:00 -> 2023-01-09 14:30:00
DEBUG       AAPL: OHLC after cleaning: 2023-01-03 09:30:00-05:00 -> 2023-01-09 09:30:00-05:00
DEBUG       AAPL: OHLC after combining events: 2023-01-03 00:00:00-05:00 -> 2023-01-09 00:00:00-05:00
DEBUG       AAPL: yfinance returning OHLC: 2023-01-03 00:00:00-05:00 -> 2023-01-09 00:00:00-05:00
DEBUG      Exiting history()
DEBUG     Exiting history()
DEBUG    Exiting download()
DEBUG         response code=200
DEBUG        Exiting _make_request()
DEBUG       Exiting get()
DEBUG       AAPL: yfinance received OHLC data: 2022-12-01 14:30:00 -> 2023-01-09 14:30:00
DEBUG       AAPL: OHLC after cleaning: 2022-12-01 09:30:00-05:00 -> 2023-01-09 09:30:00-05:00
DEBUG       AAPL: OHLC after combining events: 2022-12-01 00:00:00-05:00 -> 2023-01-09 00:00:00-05:00
DEBUG       AAPL: yfinance returning OHLC: 2022-12-01 00:00:00-05:00 -> 2023-01-09 00:00:00-05:00
DEBUG      Exiting history()
DEBUG     Exiting history()
DEBUG    Exiting download()

📊 Result 1:
  From: 2023-01-03 To: 2023-01-09
  First Close: 123.47
  Last Close:  128.49
  Rows: 5

📊 Result 2:
  From: 2022-12-01 To: 2023-01-09
  First Close: 146.41
  Last Close:  128.49
  Rows: 26

Bad data proof

📊 Result 1:
From: 2022-12-01 To: 2023-01-09
First Close: 146.41
Last Close: 128.49
Rows: 26

📊 Result 2:
From: 2022-12-01 To: 2023-01-09
First Close: 146.41
Last Close: 128.49
Rows: 26

yfinance version

0.2.64

Python version

3.13

Operating system

macOS Sequoia 15.5

[tnadav]

As a possible improvement, I was thinking of replacing the use of threads with asyncio and implementing an async version of the fetching API. This could:

• Avoid thread-safety issues like the shared _DFS dictionary.
• Enable more efficient concurrent fetching using asyncio and aiohttp.
• Integrate well with async-based applications (e.g., web servers or data pipelines).
• Still allow for backward compatibility with the current API.

I’d be happy to explore this and maybe prototype something if there’s community interest in such a direction. Would this kind of contribution be welcome?

[ValueRaider]

Just add a mutex lock? Maybe 5 LOC, much simpler.

an async version of the fetching API

Can this be a new package that sits on top of yfinance? Then yfinance focuses on being sync, except when Yahoo API is actually async https://ranaroussi.github.io/yfinance/reference/yfinance.websocket.html

[tnadav]

I think for me this is more "kill two birds with one stone" situation, where I'd also like an async version of yfinance, and as the HTTP requests are IO bound, there is no need for actual threads to make requests concurrently.
To be clear, what I suggested was a patch which replaces yfinance usage of requests with an async HTTP library such as httpx, and implement multi.py logic using asyncio tasks to get parallelism. You can't really do it without modifying the core library.
If no one is excited about it other than me there's no reason to do it...

[ValueRaider]

we had to move away from requests, anything other than curl_cffi probably won't work
https://curl-cffi.readthedocs.io/en/latest/quick_start.html#asyncio

[nadav-next]

curl_cffi seems like an awesome package and it has async interface so I don't see the conflict here

[mathieugouin]

Just chiming in that I have similar problem but with using the builtin threading module.