How to avoid a Broken Pipe error when printing a large amount of formatted data?

Question

I am trying to print a list of tuples formatted in my stdout. For this, I use the str.format method. Everything works fine, but when I pipe the output to see the first lines using the head command a IOError occurs.

Here is my code:

# creating the data
data = []$
for i in range(0,  1000):                                            
  pid = 'pid%d' % i
  uid = 'uid%d' % i
  pname = 'pname%d' % i
  data.append( (pid, uid, pname) )

# find max leghed string for each field
pids, uids, pnames = zip(*data)
max_pid = len("%s" % max( pids) )
max_uid = len("%s" % max( uids) )
max_pname = len("%s" % max( pnames) )

# my template for the formatted strings
template = "{0:%d}\t{1:%d}\t{2:%d}" % (max_pid, max_uid, max_pname)

# print the formatted output to stdout
for pid, uid, pname in data:
  print template.format(pid, uid, pname)

And here is the error I get after running the command: python myscript.py | head

Traceback (most recent call last):
  File "lala.py", line 16, in <module>
    print template.format(pid, uid, pname)
IOError: [Errno 32] Broken pipe

Can anyone help me on this?

I tried to put print in a try-except block to handle the error, but after that there was another message in the console:

close failed in file object destructor:
sys.excepthook is missing
lost sys.stderr

I also tried to flush immediately the data through a two consecutive sys.stdout.write and sys.stdout.flush calls, but nothing happend..

Conflux

This happens because head closes stdout, causing print to try and write to a closed file. What would you like to happen instead? – Refugio

Ok, thank you! I would like to avoid the printing of such messages in the console. I want to use a variation of this code to a command line tool. – Conflux

This question is a possible duplicate; see: #11423725 – Sines

Duplicate of: IOError: [Errno 32] Broken pipe when piping: `prog.py | othercmd` – Untrue

Answer 1

head reads from stdout then closes it. This causes print to fail, internally it writes to sys.stdout, now closed.

You can simply catch the IOError and exit silently:

try:
    for pid, uid, pname in data:
        print template.format(pid, uid, pname)
except IOError:
    # stdout is closed, no point in continuing
    # Attempt to close them explicitly to prevent cleanup problems:
    try:
        sys.stdout.close()
    except IOError:
        pass
    try:
        sys.stderr.close()
    except IOError:
        pass

Refugio

ok in this simple program this works, but in the command line tool sometimes works and sometimes occurs again this message: close failed in file object destructor: sys.excepthook is missing lost sys.stderr – Conflux

Haha! I found it too in some answer at stackoverflow! i tried it but it doesn't work! It seems that there is a race condition.. But I cannot figure out how to solve it.. – Conflux

ok I found it! It needs just an addition to your answer. Just put a sys.stdout.flush() right after the print statement!! But the error is what you described! I also noticed that for some reason if I put the flush out of the loop it works too.. Anyway. thank you! :) – Conflux

@ThanasisPetsas: The flush can be affected by the close too. – Refugio

I used the first version of your answer with the flush after the print and everything works great! – Conflux

Glad to have been of help then. :-) The flush may work for you because when using pipes python uses a regular file buffer instead of a line buffer, and flushing helps get your data out line by line in that case. – Refugio

Thank you very much! I have come across you answers many times and they have helped me! :) – Conflux

unfortunately this doesn't work in Python 3. Python 3 raises BrokenPipeError, but _io.TextIOWrapper catches it and emits it as a warning. So the failure isn't quite as bad but still ugly and annoying, and there's no way I can see yet to suppress it. – Simonnesimonpure

Warnings can be silenced with the warnings module; or is this not one of those? – Refugio

Answer 2

The behavior you are seeing is linked to the buffered output implementation in Python3. The problem can be avoided using the -u option or setting environmental variable PYTHONUNBUFFERED=x. See the man pages for more information on -u.

$ python2.7 testprint.py | echo

Exc: <type 'exceptions.IOError'>
$ python3.5 testprint.py | echo

Exc: <class 'BrokenPipeError'>
Exception ignored in: <_io.TextIOWrapper name='<stdout>' mode='w' encoding='UTF-8'>
BrokenPipeError: [Errno 32] Broken pipe
$ python3.5 -u testprint.py | echo

Exc: <class 'BrokenPipeError'>
$ export PYTHONUNBUFFERED=x
$ python3.5 testprint.py | echo

Exc: <class 'BrokenPipeError'>

Ureter

Yup, but I had this problem in Python2.7. I don't use Python3. – Conflux

Answer 3

In general, I try to catch the most specific error I can get away with. In this case it is BrokenPipeError:

try:
    # I usually call a function here that generates all my output:
    for pid, uid, pname in data:
        print template.format(pid, uid, pname)
except BrokenPipeError as e:
    pass  # Ignore. Something like head is truncating output.
finally:
    sys.stderr.close()

If this is at the end of execution, I find I only need to close sys.stderr. If I don't close sys.stderr, I'll get a BrokenPipeError but without a stack trace.

This seems to be the minimum fix for writing tools that output to pipelines.

Pinnatifid

Answer 4

Had this problem with Python3 and debug logging piped into head as well. If your script talks to the network or does file IO, simply dropping IOError's is not a good solution. Despite mentions here, I was not able to catch BrokenPipeError for some reason.

Found a blog post talking about restoring the default signal handler for sigpipe: http://newbebweb.blogspot.com/2012/02/python-head-ioerror-errno-32-broken.html

In short, you add the following to your script before the bulk of the output:

if log.isEnabledFor(logging.DEBUG):  # optional
    # set default handler to no-op
    from signal import signal, SIGPIPE, SIG_DFL
    signal(SIGPIPE, SIG_DFL)

This seems to happen with head, but not other programs such as grep---as mentioned head closes stdout. If you don't use head with the script often, it may not be worth worrying about.

Goldin

Great! Thanks for the explanation and the article! – Conflux