When starting a long-running Python program, I often run it under nohup so I can redirect all the output to nohup.out for easier inspection. I use nohup python test.py& to run the process in the background. Then I can use tail -f nohup.out to monitor the output from this program.

One strange issue I met is that all messages from the print() function are missing. To reproduce this issue, run the following test.py with command nohup python test.py&:

import time

n = 20
for i in range(n):
    print(f"this is loop {i}")

Then use tail -f nohup.out to monitor the output. Notice that there is no output for some time. Then, all of a sudden, all the output is printed.

However, if we directly run this program (python test.py), to our expectations, the messages are shown one at a time.

Buffering behavior of sys.stdout and sys.stderr

This has something to do with the stream buffering behavior of Python. The print() function in Python will print message to the sys.stdout stream. When we print directly to the console, in this case, the stdout stream is used interactively, it will be line-buffered, i.e., the message will be shown in the terminal once a new line is met.

If we use nohup, stdout is redirected to a file (nohup.out), it will be block-buffered in this case, i.e., only when the size of the output reaches a certain limit, will they be put into the destination (nohup.out file). So we can not see each print message in a timely manner. When the program stops, all the output is flushed to nohup.out, that is when we see those messages.

This behavior of sys.stdout and sys.stderr is documented here:

When interactive, stdout and stderr streams are line-buffered. Otherwise, they are block-buffered like regular text files.

This explains why the messages is not shown promptly if we redirect stdout to nohup.out, since stdout is block buffered in this case. If we use sufficiently large message, it will also be shown immediately since the output buffer is full.

To verify this, we modify test.py a little bit:

import time

n = 20
for i in range(n):
    print(f"{i}" + "a" * 8192)

Run it using nohup and check the output, you will find that the messages are indeed printed separately, since each message is now big enough (more than 8192 bytes).

For stderr, users may want to see the error messages immediately when they are produced even if it is redirected. In Python 3.9, the behavior of stderr is changed. The new doc is:

When interactive, the stdout stream is line-buffered. Otherwise, it is block-buffered like regular text files. The stderr stream is line-buffered in both cases.

Thus, after Python 3.9, stderr will always be line-buffered, whether the error message is displayed on the terminal or redirected to a file.

Show the message without buffering

To show the printed message immediately, we may run Python in unbuffered mode with the -u flag:

-u : force the stdout and stderr streams to be unbuffered

nohup python -u test.py

Another way is to add flush=True to print() function so that output will be flushed to its destination forcibly.

print(f"this is loop {i}", flush=True)