Conclusion

• Usually, when an error or exception is handled, the runtime may terminate. Tenacity reruns a function multiple times without terminating the runtime.

Situation

While preprocessing and running pytest for the BEIR dataset benchmark, an unfortunate situation occurred: even though I had sliced 7 datasets, it still took an entire hour.

While exploring the depths of the Ragas code to determine the cause, I discovered code using Tenacity to specify the maximum number of retries. This led me to look into Tenacity and write this article.

Usage

pip install tenacity

import tenacity

def throw_error():
    print("running...")
    raise ValueError("Errors make me stronger")

if __name__ == "__main__":
    throw_error()

When run this way, an error occurs immediately, after which the script terminates right away.

running...
Traceback (most recent call last):
...
ValueError: Errors make me stronger

As seen in the depths of Ragas, Tenacity allows you to try multiple times.

@tenacity.retry( # 데코레이터 추가
    stop=tenacity.stop_after_attempt(5), # stop 파라미터 추가
)
def throw_error():
    print("running...")
    raise ValueError("Errors make me stronger")

running... # 1번
running... # 2번
running... # 3번
running... # 4번
running... # 5번
Traceback (most recent call last):
...
tenacity.RetryError: RetryError[<Future at 0x1050b6370 state=finished raised ValueError>]

How to set the execution interval

@tenacity.retry(
    wait=tenacity.wait_fixed(10), # wait 파라미터 추가
    stop=tenacity.stop_after_attempt(5),
)
def throw_error():
    date = datetime.datetime.now().time() # 시간을 확인하기위해 추가
    print(f"time: {date}, running... ")
    raise ValueError("Errors make me stronger")

I specified tenacity.wait_fixed(10) for the wait parameter. Here, 10 means 10 seconds.

When you run the script, it executes 5 times at 10-second intervals this time.

time: 13:29:27.350657, running... 
time: 13:29:37.355707, running... 
time: 13:29:47.358180, running... 
time: 13:29:57.368775, running... 
time: 13:30:07.371236, running... 
Traceback (most recent call last):
...
tenacity.RetryError: RetryError[<Future at 0x1054425e0 state=finished raised ValueError>]

How did Ragas use it?

It set the retry count and time so that it would retry once a certain amount of time had passed.

References https://daco2020.tistory.com/520