SYNOPSIS
use File::Write::Rotate;
my $fwr = File::Write::Rotate->new(
dir => '/var/log', # required
prefix => 'myapp', # required
#suffix => '.log', # default is ''
size => 25*1024*1024, # default is 10MB, unless period is set
histories => 12, # default is 10
#buffer_size => 100, # default is none
);
# write, will write to /var/log/myapp.log, automatically rotate old log files
# to myapp.log.1 when myapp.log reaches 25MB. will keep old log files up to
# myapp.log.12.
$fwr->write("This is a line\n");
$fwr->write("This is", " another line\n");
To compressing old log files:
$fwr->compress;
This is usually done in a separate process, because it potentially
takes a long time if the files to compress are large; we are rotating
automatically in write() so doing automatic compression too would
annoyingly block writer for a potentially long time.
DESCRIPTION
This module can be used to write to file, usually for logging, that can
rotate itself. File will be opened in append mode. By default, locking
will be done to avoid conflict when there are multiple writers.
Rotation can be done by size (after a certain size is reached), by time
(daily/monthly/yearly), or both.
I first wrote this module for logging script STDERR output to files
(see Tie::Handle::FileWriteRotate).
ATTRIBUTES
buffer_size => int
Get or set buffer size. If set to a value larger than 0, then when a
write() failed, instead of dying, the message will be stored in an
internal buffer first (a regular Perl array). When the number of items
in the buffer exceeds this size, then write() will die upon failure.
Otherwise, every write() will try to flush the buffer.
Can be used for example when a program runs as superuser/root then
temporarily drops privilege to a normal user. During this period,
logging can fail because the program cannot lock the lock file or write
to the logging directory. Before dropping privilege, the program can
set buffer_size to some larger-than-zero value to hold the messages
emitted during dropping privilege. The next write() as the
superuser/root will succeed and flush the buffer to disk (provided
there is no other error condition, of course).
path => str (ro)
Current file's path.
handle => (ro)
Current file handle. You should not use this directly, but use write()
instead. This attribute is provided for special circumstances (e.g. in
hooks, see example in the hook section).
hook_before_write => code
Will be called by write() before actually writing to filehandle (but
after locking is done). Code will be passed ($self, \@msgs, $fh) where
@msgs is an array of strings to be written (the contents of buffer, if
any, plus arguments passed to write()) and $fh is the filehandle.
hook_before_rotate => code
Will be called by the rotating routine before actually doing rotating.
Code will be passed ($self).
This can be used to write a footer to the end of each file, e.g.:
# hook_before_rotate
my ($self) = @_;
my $fh = $self->handle;
print $fh "Some footer\n";
Since this hook is indirectly called by write(), locking is already
done.
hook_after_rotate => code
Will be called by the rotating routine after the rotating process. Code
will be passed ($self, \@renamed, \@deleted) where @renamed is array of
new filenames that have been renamed, @deleted is array of new
filenames that have been deleted.
hook_after_create => code
Will be called by after a new file is created. Code will be passed
($self).
This hook can be used to write a header to each file, e.g.:
# hook_after_create
my ($self) = @_;
my $fh $self->handle;
print $fh "header\n";
Since this is called indirectly by write(), locking is also already
done.
binmode => str
METHODS
$obj = File::Write::Rotate->new(%args)
Create new object. Known arguments:
* dir => STR (required)
Directory to put the files in.
* prefix => STR (required)
Name of files. The files will be named like the following:
<prefix><period><suffix><rotate_suffix>
<period> will only be given if the period argument is set. If period
is set to yearly, <period> will be YYYY (4-digit year). If period is
monthly, <period> will be YYYY-MM (4-digit year and 2-digit month).
If period is daily, <period> will be YYYY-MM-DD (4-digit year,
2-digit month, and 2-digit day).
<rotate_suffix> is either empty string for current file; or .1, .2
and so on for rotated files. .1 is the most recent rotated file, .2
is the next most recent, and so on.
An example, with prefix set to myapp:
myapp # current file
myapp.1 # most recently rotated
myapp.2 # the next most recently rotated
With prefix set to myapp, period set to monthly, suffix set to .log:
myapp.2012-12.log # file name for december 2012
myapp.2013-01.log # file name for january 2013
Like previous, but additionally with size also set (which will also
rotate each period file if it exceeds specified size):
myapp.2012-12.log # file(s) for december 2012
myapp.2012-12.log.1
myapp.2012-12.log.2
myapp.2013-01.log # file(s) for january 2013
All times will use local time, so you probably want to set TZ
environment variable or equivalent methods to set time zone.
* suffix => STR (default: '')
Suffix to give to file names, usually file extension like .log. See
prefix for more details.
If you use a yearly period, setting suffix is advised to avoid
ambiguity with rotate suffix (for example, is myapp.2012 the current
file for year 2012 or file with 2012 rotate suffix?)
* size => INT (default: 10*1024*1024)
Maximum file size, in bytes, before rotation is triggered. The
default is 10MB (10*1024*1024) if period is not set. If period is
set, no default for size is provided, which means files will not be
rotated for size (only for period).
* period => STR
Can be set to either daily, monthly, or yearly. If set, will
automatically rotate after period change. See prefix for more
details.
* histories => INT (default: 10)
Number of rotated files to keep. After the number of files exceeds
this, the oldest one will be deleted. 0 means not to keep any
history, 1 means to only keep .1 file, and so on.
* buffer_size => INT (default: 0)
Set initial value of buffer. See the buffer_size attribute for more
information.
* lock_mode => STR (default: 'write')
Can be set to either none, write, or exclusive. none disables locking
and increases write performance, but should only be used when there
is only one writer. write acquires and holds the lock for each write.
exclusive acquires the lock at object creation and holds it until the
the object is destroyed.
Lock file is named <prefix>.lck. Will wait for up to 1 minute to
acquire lock, will die if failed to acquire lock.
* hook_before_write => CODE
* hook_before_rotate => CODE
* hook_after_rotate => CODE
* hook_after_create => CODE
See "ATTRIBUTES".
* buffer_size => int
* rotate_probability => float (between 0 < x < 1)
If set, instruct to only check for rotation under a certain
probability, for example if value is set to 0.1 then will only check
for rotation 10% of the time.
lock_file_path => STR
Returns a string representing the complete pathname to the lock file,
based on dir and prefix attributes.
$fwr->write(@args)
Write to file. Will automatically rotate file if period changes or file
size exceeds specified limit. When rotating, will only keep a specified
number of histories and delete the older ones.
Does not append newline so you'll have to do it yourself.
$fwr->compress
Compress old rotated files and remove the uncompressed originals.
Currently uses IO::Compress::Gzip to do the compression. Extension
given to compressed file is .gz.
Will not lock writers, but will create <prefix>-compress.pid PID file
to prevent multiple compression processes running and to signal the
writers to postpone rotation.
After compression is finished, will remove the PID file, so rotation
can be done again on the next write() if necessary.
FAQ
Why use autorotating file?
Mainly convenience and low maintenance. You no longer need a separate
rotator process like the Unix logrotate utility (which when
accidentally disabled or misconfigured will cause your logs to stop
being rotated and grow indefinitely).
What is the downside of using FWR (and LDFR)?
Mainly (significant) performance overhead. At (almost) every write(),
FWR needs to check file sizes and/or dates for rotation. Under default
configuration (where lock_mode is write), it also performs locking on
each write() to make it safe to use with multiple processes. Below is a
casual benchmark to give a sense of the overhead, tested on my Core
i5-2400 3.1GHz desktop:
Writing lines in the size of ~ 200 bytes, raw writing to disk (SSD) has
the speed of around 3.4mil/s, while using FWR it goes down to around
~13k/s. Using lock_mode none or exclusive, the speed is ~52k/s.
However, this is not something you'll notice or need to worry about
unless you're writing near that speed.
If you need more speed, you can try setting rotate_probability which
will cause FWR to only check for rotation probabilistically, e.g. if
you set this to 0.1 then checks will only be done in about 1 of 10
writes. This can significantly reduce the overhead and increase write
speed several times (e.g. 5-8 times), but understand that this will
make the writes "overflow" a bit, e.g. file sizes will exceed for a bit
if you do size-based rotation. More suitable if you only do size-based
rotation since it is usually okay to exceed sizes for a bit.
SEE ALSO
Log::Dispatch::FileRotate, which inspires this module. Differences
between File::Write::Rotate (FWR) and Log::Dispatch::FileRotate (LDFR)
are as follows:
* FWR is not part of the Log::Dispatch family.
This makes FWR more general to use.
For using together with Log::Dispatch/Log4perl, I have also written
Log::Dispatch::FileWriteRotate which is a direct (although not a
perfect drop-in) replacement for Log::Dispatch::FileRotate.
* Secondly, FWR does not use Date::Manip.
Date::Manip is relatively large (loading Date::Manip 6.37 equals to
loading 34 files and ~ 22k lines; while FWR itself is only < 1k
lines!)
As a consequence of this, FWR does not support DatePattern; instead,
FWR replaces it with a simple daily/monthly/yearly period.
* And lastly, FWR supports compressing and rotating compressed old
files.
Using separate processes like the Unix logrotate utility means having
to deal with yet another race condition. FWR takes care of that for
you (see the compress() method). You also have the option to do file
compression in the same script/process if you want, which is
convenient.
There is no significant overhead difference between FWR and LDFR (FWR
is slightly faster than LDFR on my testing).
Tie::Handle::FileWriteRotate and Log::Dispatch::FileWriteRotate, which
use this module.