Where does print go on a worker?

[jsignell]

I have been staring at this for a while and can't figure it out. When I call print in a task within a distributed worker, where does that print happen?

When I use a LocalCluster from Jupyter it prints in the terminal where I started my Jupyter from. If I want it to instead print to my logs I can do:

from dask.distributed import Client
from distributed.worker import logger

client = Client()

def log_it():    
    logger.info("This will output to the logs!")
    
client.run(log_it)

What I'd like is for this to print in my worker logs as well.

from dask.distributed import Client

client = Client()

def print_it():    
    print("Print this in the logs!")
    
client.run(print_it)

[quasiben]

When you print it should go to stdout of the worker. Those won't typically be reflected in the logs captured by dask for example at say: http://localhost:8787/info/logs/tcp%3A%2F%2F127.0.0.1%3A53082.html. I think what you have down is correct but maybe there is a config setting in place for the worker ? Could you try with info/debug/error/warn ?

xref: https://docs.dask.org/en/latest/debugging.html#logs for configuring client/scheduler/worker logs

[jrbourbeau]

FWIW there was a discussion about capturing stdout/stderr on workers and redirecting to into Dask's logging system at some point (xref dask/distributed#2033)

[jsignell]

Ah thanks yeah. I came across that issue in my search, but neglected to mention it. I had trouble telling if the information in that issue was uptodate.

[jsignell]

@quasiben I tried setting worker logging to debug and it is more verbose, but still doesn't capture the print output.

[jsignell]

I opened a PR to discuss this more: dask/distributed#4502