I would like to make tqdm works when executed in a cloud env where logs are propagated as text to external services. I am mostly talking about Kubernetes here, but I guess it also apply to any other container orchestration tool.

Consider this simple snippet using tqdm 4.64:

import sys
from tqdm import tqdm
from time import sleep

for i in tqdm(range(10)):
    sleep(5)
    sys.stderr.write("\n")

When removing sys.stderr.write("\n") then the tqdm progress bar only display at the very end of the loop in one shot to 100%. It obviously defeats the purpose of the progress bar.

Adding sys.stderr.write("\n") trigger a log flush (or propagation) to wherever the logs are sent (Kubernetes console for example) and so the progress bar is displayed at regular interval.

This simple snippet is actually a workaround to make tqdm works with Kubernetes, but it does not reflect the complexity of a given stack where things like progress bar can be deeply nested into the codebase and so not "tunable" when executing a workload.

I have two questions here:

• Is there is actually another more "automatic" workaround than sys.stderr.write("\n")?
• Would you be willing to add to tqdm(range(10), ...) a new arg that can trigger this sys.stderr.write("\n") whenever the bar must be updated? We could eventually imagine auto-detecting (if possible) Kubernetes or simply setting a boolean flag.
  • On this it seems important to me to add a new tqdm arg instead of a more complex solution since in a deeply nested codebase, tqdm args is often the only thing you have control on. Auto-detection would be even better, if that something easily doable (I can dig into this if needed).