mylar/lib/ConcurrentLogHandler/cloghandler.py

350 lines
15 KiB
Python
Raw Normal View History

# Copyright 2013 Lowell Alleman
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may not
# use this file except in compliance with the License. You may obtain a copy
# of the License at http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
""" cloghandler.py: A smart replacement for the standard RotatingFileHandler
ConcurrentRotatingFileHandler: This class is a log handler which is a drop-in
replacement for the python standard log handler 'RotateFileHandler', the primary
difference being that this handler will continue to write to the same file if
the file cannot be rotated for some reason, whereas the RotatingFileHandler will
strictly adhere to the maximum file size. Unfortunately, if you are using the
RotatingFileHandler on Windows, you will find that once an attempted rotation
fails, all subsequent log messages are dropped. The other major advantage of
this module is that multiple processes can safely write to a single log file.
To put it another way: This module's top priority is preserving your log
records, whereas the standard library attempts to limit disk usage, which can
potentially drop log messages. If you are trying to determine which module to
use, there are number of considerations: What is most important: strict disk
space usage or preservation of log messages? What OSes are you supporting? Can
you afford to have processes blocked by file locks?
Concurrent access is handled by using file locks, which should ensure that log
messages are not dropped or clobbered. This means that a file lock is acquired
and released for every log message that is written to disk. (On Windows, you may
also run into a temporary situation where the log file must be opened and closed
for each log message.) This can have potentially performance implications. In my
testing, performance was more than adequate, but if you need a high-volume or
low-latency solution, I suggest you look elsewhere.
This module currently only support the 'nt' and 'posix' platforms due to the
usage of the portalocker module. I do not have access to any other platforms
for testing, patches are welcome.
See the README file for an example usage of this module.
This module supports Python 2.6 and later.
"""
__version__ = '0.9.1'
__revision__ = 'lowell87@gmail.com-20130711022321-doutxl7zyzuwss5a 2013-07-10 22:23:21 -0400 [0]'
__author__ = "Lowell Alleman"
__all__ = [
"ConcurrentRotatingHandler",
]
import os
import sys
from random import randint
from logging import Handler, LogRecord
from logging.handlers import BaseRotatingHandler
try:
import codecs
except ImportError:
codecs = None
# Question/TODO: Should we have a fallback mode if we can't load portalocker /
# we should still be better off than with the standard RotattingFileHandler
# class, right? We do some rename checking... that should prevent some file
# clobbering that the builtin class allows.
# sibling module than handles all the ugly platform-specific details of file locking
from portalocker import lock, unlock, LOCK_EX, LOCK_NB, LockException
# Workaround for handleError() in Python 2.7+ where record is written to stderr
class NullLogRecord(LogRecord):
def __init__(self):
pass
def __getattr__(self, attr):
return None
class ConcurrentRotatingFileHandler(BaseRotatingHandler):
"""
Handler for logging to a set of files, which switches from one file to the
next when the current file reaches a certain size. Multiple processes can
write to the log file concurrently, but this may mean that the file will
exceed the given size.
"""
def __init__(self, filename, mode='a', maxBytes=0, backupCount=0,
encoding=None, debug=True, delay=0):
"""
Open the specified file and use it as the stream for logging.
By default, the file grows indefinitely. You can specify particular
values of maxBytes and backupCount to allow the file to rollover at
a predetermined size.
Rollover occurs whenever the current log file is nearly maxBytes in
length. If backupCount is >= 1, the system will successively create
new files with the same pathname as the base file, but with extensions
".1", ".2" etc. appended to it. For example, with a backupCount of 5
and a base file name of "app.log", you would get "app.log",
"app.log.1", "app.log.2", ... through to "app.log.5". The file being
written to is always "app.log" - when it gets filled up, it is closed
and renamed to "app.log.1", and if files "app.log.1", "app.log.2" etc.
exist, then they are renamed to "app.log.2", "app.log.3" etc.
respectively.
If maxBytes is zero, rollover never occurs.
On Windows, it is not possible to rename a file that is currently opened
by another process. This means that it is not possible to rotate the
log files if multiple processes is using the same log file. In this
case, the current log file will continue to grow until the rotation can
be completed successfully. In order for rotation to be possible, all of
the other processes need to close the file first. A mechanism, called
"degraded" mode, has been created for this scenario. In degraded mode,
the log file is closed after each log message is written. So once all
processes have entered degraded mode, the net rotation attempt should
be successful and then normal logging can be resumed. Using the 'delay'
parameter may help reduce contention in some usage patterns.
This log handler assumes that all concurrent processes logging to a
single file will are using only this class, and that the exact same
parameters are provided to each instance of this class. If, for
example, two different processes are using this class, but with
different values for 'maxBytes' or 'backupCount', then odd behavior is
expected. The same is true if this class is used by one application, but
the RotatingFileHandler is used by another.
"""
# Absolute file name handling done by FileHandler since Python 2.5
BaseRotatingHandler.__init__(self, filename, mode, encoding, delay)
self.delay = delay
self._rotateFailed = False
self.maxBytes = maxBytes
self.backupCount = backupCount
self._open_lockfile()
# For debug mode, swap out the "_degrade()" method with a more a verbose one.
if debug:
self._degrade = self._degrade_debug
def _open_lockfile(self):
# Use 'file.lock' and not 'file.log.lock' (Only handles the normal "*.log" case.)
if self.baseFilename.endswith(".log"):
lock_file = self.baseFilename[:-4]
else:
lock_file = self.baseFilename
lock_file += ".lock"
self.stream_lock = open(lock_file,"w")
def _open(self, mode=None):
"""
Open the current base file with the (original) mode and encoding.
Return the resulting stream.
Note: Copied from stdlib. Added option to override 'mode'
"""
if mode is None:
mode = self.mode
if self.encoding is None:
stream = open(self.baseFilename, mode)
else:
stream = codecs.open(self.baseFilename, mode, self.encoding)
return stream
def _close(self):
""" Close file stream. Unlike close(), we don't tear anything down, we
expect the log to be re-opened after rotation."""
if self.stream:
try:
if not self.stream.closed:
# Flushing probably isn't technically necessary, but it feels right
self.stream.flush()
self.stream.close()
finally:
self.stream = None
def acquire(self):
""" Acquire thread and file locks. Re-opening log for 'degraded' mode.
"""
# handle thread lock
Handler.acquire(self)
# Issue a file lock. (This is inefficient for multiple active threads
# within a single process. But if you're worried about high-performance,
# you probably aren't using this log handler.)
if self.stream_lock:
# If stream_lock=None, then assume close() was called or something
# else weird and ignore all file-level locks.
if self.stream_lock.closed:
# Daemonization can close all open file descriptors, see
# https://bugzilla.redhat.com/show_bug.cgi?id=952929
# Try opening the lock file again. Should we warn() here?!?
try:
self._open_lockfile()
except Exception:
self.handleError(NullLogRecord())
# Don't try to open the stream lock again
self.stream_lock = None
return
lock(self.stream_lock, LOCK_EX)
# Stream will be opened as part by FileHandler.emit()
def release(self):
""" Release file and thread locks. If in 'degraded' mode, close the
stream to reduce contention until the log files can be rotated. """
try:
if self._rotateFailed:
self._close()
except Exception:
self.handleError(NullLogRecord())
finally:
try:
if self.stream_lock and not self.stream_lock.closed:
unlock(self.stream_lock)
except Exception:
self.handleError(NullLogRecord())
finally:
# release thread lock
Handler.release(self)
def close(self):
"""
Close log stream and stream_lock. """
try:
self._close()
if not self.stream_lock.closed:
self.stream_lock.close()
finally:
self.stream_lock = None
Handler.close(self)
def _degrade(self, degrade, msg, *args):
""" Set degrade mode or not. Ignore msg. """
self._rotateFailed = degrade
del msg, args # avoid pychecker warnings
def _degrade_debug(self, degrade, msg, *args):
""" A more colorful version of _degade(). (This is enabled by passing
"debug=True" at initialization).
"""
if degrade:
if not self._rotateFailed:
sys.stderr.write("Degrade mode - ENTERING - (pid=%d) %s\n" %
(os.getpid(), msg % args))
self._rotateFailed = True
else:
if self._rotateFailed:
sys.stderr.write("Degrade mode - EXITING - (pid=%d) %s\n" %
(os.getpid(), msg % args))
self._rotateFailed = False
def doRollover(self):
"""
Do a rollover, as described in __init__().
"""
self._close()
if self.backupCount <= 0:
# Don't keep any backups, just overwrite the existing backup file
# Locking doesn't much matter here; since we are overwriting it anyway
self.stream = self._open("w")
return
try:
# Determine if we can rename the log file or not. Windows refuses to
# rename an open file, Unix is inode base so it doesn't care.
# Attempt to rename logfile to tempname: There is a slight race-condition here, but it seems unavoidable
tmpname = None
while not tmpname or os.path.exists(tmpname):
tmpname = "%s.rotate.%08d" % (self.baseFilename, randint(0,99999999))
try:
# Do a rename test to determine if we can successfully rename the log file
os.rename(self.baseFilename, tmpname)
except (IOError, OSError):
exc_value = sys.exc_info()[1]
self._degrade(True, "rename failed. File in use? "
"exception=%s", exc_value)
return
# Q: Is there some way to protect this code from a KeboardInterupt?
# This isn't necessarily a data loss issue, but it certainly does
# break the rotation process during stress testing.
# There is currently no mechanism in place to handle the situation
# where one of these log files cannot be renamed. (Example, user
# opens "logfile.3" in notepad); we could test rename each file, but
# nobody's complained about this being an issue; so the additional
# code complexity isn't warranted.
for i in range(self.backupCount - 1, 0, -1):
sfn = "%s.%d" % (self.baseFilename, i)
dfn = "%s.%d" % (self.baseFilename, i + 1)
if os.path.exists(sfn):
#print "%s -> %s" % (sfn, dfn)
if os.path.exists(dfn):
os.remove(dfn)
os.rename(sfn, dfn)
dfn = self.baseFilename + ".1"
if os.path.exists(dfn):
os.remove(dfn)
os.rename(tmpname, dfn)
#print "%s -> %s" % (self.baseFilename, dfn)
self._degrade(False, "Rotation completed")
finally:
# Re-open the output stream, but if "delay" is enabled then wait
# until the next emit() call. This could reduce rename contention in
# some usage patterns.
if not self.delay:
self.stream = self._open()
def shouldRollover(self, record):
"""
Determine if rollover should occur.
For those that are keeping track. This differs from the standard
library's RotatingLogHandler class. Because there is no promise to keep
the file size under maxBytes we ignore the length of the current record.
"""
del record # avoid pychecker warnings
# Is stream is not yet open, skip rollover check. (Check will occur on
# next message, after emit() calls _open())
if self.stream is None:
return False
if self._shouldRollover():
# If some other process already did the rollover (which is possible
# on Unix) the file our stream may now be named "log.1", thus
# triggering another rollover. Avoid this by closing and opening
# "log" again.
self._close()
self.stream = self._open()
return self._shouldRollover()
return False
def _shouldRollover(self):
if self.maxBytes > 0: # are we rolling over?
self.stream.seek(0, 2) #due to non-posix-compliant Windows feature
if self.stream.tell() >= self.maxBytes:
return True
else:
self._degrade(False, "Rotation done or not needed at this time")
return False
# Publish this class to the "logging.handlers" module so that it can be use
# from a logging config file via logging.config.fileConfig().
import logging.handlers
logging.handlers.ConcurrentRotatingFileHandler = ConcurrentRotatingFileHandler