I'd like to make the output of tail -F or something similar available to me in Python without blocking or locking. I've found some really old code to do that here, but I'm thinking there must be a better way or a library to do the same thing by now. Anyone know of one?
Ideally, I'd have something like tail.getNewData() that I could call every time I wanted more data.
So, this is coming quite late, but I ran into the same problem again, and there's a much better solution now. Just use pygtail:
Pygtail reads log file lines that have not been read. It will even handle log files that have been rotated. Based on logcheck's logtail2 (http://logcheck.org)
If you are on linux (as windows does not support calling select on files) you can use the subprocess module along with the select module.
import time
import subprocess
import select
f = subprocess.Popen(['tail','-F',filename],\
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
p = select.poll()
p.register(f.stdout)
while True:
if p.poll(1):
print f.stdout.readline()
time.sleep(1)
This polls the output pipe for new data and prints it when it is available. Normally the time.sleep(1) and print f.stdout.readline() would be replaced with useful code.
You can use the subprocess module without the extra select module calls.
import subprocess
f = subprocess.Popen(['tail','-F',filename],\
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
while True:
line = f.stdout.readline()
print line
This will also print new lines as they are added, but it will block until the tail program is closed, probably with f.kill().
f.stdout is a pipe, not a file (but I believe Windows is still incapable of using select on it). –
Smutch print line, use sys.stdout.write(line) to take care of extra newlines that print will insert. –
Niello .strip() would also remove leading whitespace that might be significant. –
Picard .readline() keeps the newline, and print adds a new one, it should be easy to fix by using sys.stdout.write() instead of print. –
Picard if p.poll(1): with a loop that calls f.stdout.readline() until nothing is returned seems to work. Here is a gist with these modifications: gist.github.com/dwaltrip/bd3321880180f556ba0f9d1c4962b6f7 –
Cadena line = f.stdout.readline() always waits until the next line is read, if the file is present. It seems to be a non-blocking wait, though (cpu time is very short), so this might not actually be an issue. However, line = f.stdout.readline() does instantaneously return a bytes string of length 0 if the file is missing. –
Cadena print statement of course). also changed line = f.stdout.readline() to line = str(f.stdout.readline(), 'utf-8').strip() to clean up the print. –
Caruncle Using the sh module (pip install sh):
from sh import tail
# runs forever
for line in tail("-f", "/var/log/some_log_file.log", _iter=True):
print(line)
[update]
Since sh.tail with _iter=True is a generator, you can:
import sh
tail = sh.tail("-f", "/var/log/some_log_file.log", _iter=True)
Then you can "getNewData" with:
new_data = tail.next()
Note that if the tail buffer is empty, it will block until there is more data (from your question it is not clear what you want to do in this case).
[update]
This works if you replace -f with -F, but in Python it would be locking. I'd be more interested in having a function I could call to get new data when I want it, if that's possible. – Eli
A container generator placing the tail call inside a while True loop and catching eventual I/O exceptions will have almost the same effect of -F.
def tail_F(some_file):
while True:
try:
for line in sh.tail("-f", some_file, _iter=True):
yield line
except sh.ErrorReturnCode_1:
yield None
If the file becomes inaccessible, the generator will return None. However it still blocks until there is new data if the file is accessible. It remains unclear for me what you want to do in this case.
Raymond Hettinger approach seems pretty good:
def tail_F(some_file):
first_call = True
while True:
try:
with open(some_file) as input:
if first_call:
input.seek(0, 2)
first_call = False
latest_data = input.read()
while True:
if '\n' not in latest_data:
latest_data += input.read()
if '\n' not in latest_data:
yield ''
if not os.path.isfile(some_file):
break
continue
latest_lines = latest_data.split('\n')
if latest_data[-1] != '\n':
latest_data = latest_lines[-1]
else:
latest_data = input.read()
for line in latest_lines[:-1]:
yield line + '\n'
except IOError:
yield ''
This generator will return '' if the file becomes inaccessible or if there is no new data.
[update]
The second to last answer circles around to the top of the file it seems whenever it runs out of data. – Eli
I think the second will output the last ten lines whenever the tail process ends, which with -f is whenever there is an I/O error. The tail --follow --retry behavior is not far from this for most cases I can think of in unix-like environments.
Perhaps if you update your question to explain what is your real goal (the reason why you want to mimic tail --retry), you will get a better answer.
The last answer does not actually follow the tail and merely reads what's available at run time. – Eli
Of course, tail will display the last 10 lines by default... You can position the file pointer at the end of the file using file.seek, I will left a proper implementation as an exercise to the reader.
IMHO the file.read() approach is far more elegant than a subprocess based solution.
tail call inside a while True loop and catching eventual I/O exceptions will have the same effect of -F. –
Jablonski file.read() approach? tail properly handles showing the last 10 lines of the file (even if the lines are huge), reading new lines forever, waking up when new lines arrive (in a platform-dependent fashion), and opening new files when needed. In a word, the utility is quite well-designed for what it is meant to do -- reimplementing it does not seem nearly as elegant. (I will, however, admit that the sh module is pretty nifty.) –
Smutch tail -F | his_application. If there are several logfiles to be processed, having one instance for each file is simpler and safer than having a single process with several threads - there is no good reason to complicate the design adding several complex modules –
Jablonski readline would block until a newline, but you are right, readline is the right method for this job. –
Jablonski I am adapting Ijaz Ahmad Khan's answer to only yield lines when they are completely written (lines end with a newline char) gives a pythonic solution with no external dependencies:
import time
from typing import Iterator
def follow(file, sleep_sec=0.1) -> Iterator[str]:
""" Yield each line from a file as they are written.
`sleep_sec` is the time to sleep after empty reads. """
line = ''
while True:
tmp = file.readline()
if tmp is not None and tmp != "":
line += tmp
if line.endswith("\n"):
yield line
line = ''
elif sleep_sec:
time.sleep(sleep_sec)
if __name__ == '__main__':
with open("test.txt", 'r') as file:
for line in follow(file):
print(line, end='')
tail -f. Upvoted. –
Threepence sh and tail –
Fresh -> Iterator[str] and the function ought to work –
Puffer if tmp is not None to if tmp != "". –
Cletuscleve The only portable way to tail -f a file appears to be, in fact, to read from it and retry (after a sleep) if the read returns 0. The tail utilities on various platforms use platform-specific tricks (e.g. kqueue on BSD) to efficiently tail a file forever without needing sleep.
Therefore, implementing a good tail -f purely in Python is probably not a good idea, since you would have to use the least-common-denominator implementation (without resorting to platform-specific hacks). Using a simple subprocess to open tail -f and iterating through the lines in a separate thread, you can easily implement a non-blocking tail operation in Python.
Example implementation:
import threading, Queue, subprocess
tailq = Queue.Queue(maxsize=10) # buffer at most 100 lines
def tail_forever(fn):
p = subprocess.Popen(["tail", "-f", fn], stdout=subprocess.PIPE)
while 1:
line = p.stdout.readline()
tailq.put(line)
if not line:
break
threading.Thread(target=tail_forever, args=(fn,)).start()
print tailq.get() # blocks
print tailq.get_nowait() # throws Queue.Empty if there are no lines to read
tail -F into it. I can't see why adding the complexity of threading, Queue and subprocess will result in any advantage over the traditional approach. –
Jablonski tail uses a combination of inotify and select on Linux; see the source code: github.com/coreutils/coreutils/blob/master/src/tail.c#L1453 –
Smutch tail -f efficient: Which it was! Could see the lines come in as they arrived at the NAS. The need was for low latency, and this worked. (The writes were coming from a VM writing a virtual com port to a raw file on the NAS, logging data from a scientific instrument). –
Tupelo All the answers that use tail -f are not pythonic.
Here is the pythonic way: ( using no external tool or library)
def follow(thefile):
while True:
line = thefile.readline()
if not line or not line.endswith('\n'):
time.sleep(0.1)
continue
yield line
if __name__ == '__main__':
logfile = open("run/foo/access-log","r")
loglines = follow(logfile)
for line in loglines:
print(line, end='')
So, this is coming quite late, but I ran into the same problem again, and there's a much better solution now. Just use pygtail:
Pygtail reads log file lines that have not been read. It will even handle log files that have been rotated. Based on logcheck's logtail2 (http://logcheck.org)
Ideally, I'd have something like tail.getNewData() that I could call every time I wanted more data
We've already got one and itsa very nice. Just call f.read() whenever you want more data. It will start reading where the previous read left off and it will read through the end of the data stream:
f = open('somefile.log')
p = 0
while True:
f.seek(p)
latest_data = f.read()
p = f.tell()
if latest_data:
print latest_data
print str(p).center(10).center(80, '=')
For reading line-by-line, use f.readline(). Sometimes, the file being read will end with a partially read line. Handle that case with f.tell() finding the current file position and using f.seek() for moving the file pointer back to the beginning of the incomplete line. See this ActiveState recipe for working code.
str(p).center(10).center(80, '=') –
Seersucker You could use the 'tailer' library: https://pypi.python.org/pypi/tailer/
It has an option to get the last few lines:
# Get the last 3 lines of the file
tailer.tail(open('test.txt'), 3)
# ['Line 9', 'Line 10', 'Line 11']
And it can also follow a file:
# Follow the file as it grows
for line in tailer.follow(open('test.txt')):
print line
If one wants tail-like behaviour, that one seems to be a good option.
follow() the same file after it's removed / recreated, so didn't work for me :/ –
Regular make-all-the-things-work-by-themselves instead –
Philippe Another option is the tailhead library that provides both Python versions of of tail and head utilities and API that can be used in your own module.
Originally based on the tailer module, its main advantage is the ability to follow files by path i.e. it can handle situation when file is recreated. Besides, it has some bug fixes for various edge cases.
Python is "batteries included" - it has a nice solution for it: https://pypi.python.org/pypi/pygtail
Reads log file lines that have not been read. Remembers where it finished last time, and continues from there.
import sys
from pygtail import Pygtail
for line in Pygtail("some.log"):
sys.stdout.write(line)
If you are on linux you implement a non-blocking implementation in python in the following way.
import subprocess
subprocess.call('xterm -title log -hold -e \"tail -f filename\"&', shell=True, executable='/bin/csh')
print "Done"
A simple tail function from pypi app tailread
You Can use it also via pip install tailread
Recommended for tail access of large files.
from io import BufferedReader
def readlines(bytesio, batch_size=1024, keepends=True, **encoding_kwargs):
'''bytesio: file path or BufferedReader
batch_size: size to be processed
'''
path = None
if isinstance(bytesio, str):
path = bytesio
bytesio = open(path, 'rb')
elif not isinstance(bytesio, BufferedReader):
raise TypeError('The first argument to readlines must be a file path or a BufferedReader')
bytesio.seek(0, 2)
end = bytesio.tell()
buf = b""
for p in reversed(range(0, end, batch_size)):
bytesio.seek(p)
lines = []
remain = min(end-p, batch_size)
while remain > 0:
line = bytesio.readline()[:remain]
lines.append(line)
remain -= len(line)
cut, *parsed = lines
for line in reversed(parsed):
if buf:
line += buf
buf = b""
if encoding_kwargs:
line = line.decode(**encoding_kwargs)
yield from reversed(line.splitlines(keepends))
buf = cut + buf
if path:
bytesio.close()
if encoding_kwargs:
buf = buf.decode(**encoding_kwargs)
yield from reversed(buf.splitlines(keepends))
for line in readlines('access.log', encoding='utf-8', errors='replace'):
print(line)
if 'line 8' in line:
break
# line 11
# line 10
# line 9
# line 8
You can also use 'AWK' command.
See more at: http://www.unix.com/shell-programming-scripting/41734-how-print-specific-lines-awk.html
awk can be used to tail last line, last few lines or any line in a file.
This can be called from python.
# -*- coding:utf-8 -*-
import sys
import time
class Tail():
def __init__(self, file_name, callback=sys.stdout.write):
self.file_name = file_name
self.callback = callback
def follow(self, n=10):
try:
# 打开文件
with open(self.file_name, 'r', encoding='UTF-8') as f:
# with open(self.file_name,'rb') as f:
self._file = f
self._file.seek(0, 2)
# 存储文件的字符长度
self.file_length = self._file.tell()
# 打印最后10行
self.showLastLine(n)
# 持续读文件 打印增量
while True:
line = self._file.readline()
if line:
self.callback(line)
time.sleep(1)
except Exception as e:
print('打开文件失败，囧，看看文件是不是不存在，或者权限有问题')
print(e)
def showLastLine(self, n):
# 一行大概100个吧 这个数改成1或者1000都行
len_line = 100
# n默认是10，也可以follow的参数传进来
read_len = len_line * n
# 用last_lines存储最后要处理的内容
while True:
# 如果要读取的1000个字符，大于之前存储的文件长度
# 读完文件，直接break
if read_len > self.file_length:
self._file.seek(0)
last_lines = self._file.read().split('\n')[-n:]
break
# 先读1000个 然后判断1000个字符里换行符的数量
self._file.seek(-read_len, 2)
last_words = self._file.read(read_len)
# count是换行符的数量
count = last_words.count('\n')
if count >= n:
# 换行符数量大于10 很好处理，直接读取
last_lines = last_words.split('\n')[-n:]
break
# 换行符不够10个
else:
# break
# 不够十行
# 如果一个换行符也没有，那么我们就认为一行大概是100个
if count == 0:
len_perline = read_len
# 如果有4个换行符，我们认为每行大概有250个字符
else:
len_perline = read_len / count
# 要读取的长度变为2500，继续重新判断
read_len = len_perline * n
for line in last_lines:
self.callback(line + '\n')
if __name__ == '__main__':
py_tail = Tail('test.txt')
py_tail.follow(1)
© 2022 - 2024 — McMap. All rights reserved.
subprocess.call(["tail", "-F", filename])– Kegan-Fpart. – Nutriliteget_new_datamethod (PEP-8 name) need to return all data since the last call, or just the current tail (possibly losing some data)? – Drabbletail -Fis meant for reading log files which can be rotated from time to time, so it's not surprising that it would print out all lines from the beginning. – Smutchtail -Finto it. I guessed his reason to implement it in Python was to ditch the dependency on external commands, but seems like I was wrong. – Jablonski-f) tail. – Smutchtail -F filename | python script.pyas I didn't need stdin for any other purpose and that gave best performance – Michaud