Log rotation, no code change needed

This post shows you how to rotate old logs from your application. There is no change to application code. There is no specialized logging library or framework needed. It works for any language, on standard Unix platform.

What is log rotation?

Log files from applications can fill up disks over time. To prevent this, they are periodically truncated at a manageable size. The older logs are gradually removed from the system. This is known as “log rotation”.

For example, a 35 MB app.log can be split as follows:

  • 5 MB app.log
  • 10 MB app.log.1 (older)
  • 10 MB app.log.2 (older)
  • 10 MB app.log.3 (oldest)

When app.log grows to 10 MB and log rotation is invoked, app.log.3 is deleted. app.log.2 becomes app.log.3, app.log.1 becomes app.log.2, app.log becomes app.log.1. An empty app.log is created. The application opens this file and writes to it. This is why the application needs to be aware of log rotation: because it has to open the new, empty log file and write to it.

Logging frameworks like log4j (Java) and poco (python) can rotate logs for your application.

Log rotation with Unix utilities

We will use syslog, newsyslog, and logger utilities that come standard on FreeBSD. On Linux, instead of newsyslog, you’d use logrotate.

Here is the broad idea:

  1. Application logs to standard output (terminal).
  2. We pipe the output to syslog server.
  3. Syslog redirects the output to an application-specific log file.
  4. Periodically, we run a command to rotate old logs.

Logging to a file

I have a utility script, startproc.sh, that starts an executable program. Here is startproc.sh:

# usage: startproc.sh <program_name>
PROGNAME=`basename $PROG`
$PROG $* 2>&1 | logger -i -t $PROGNAME -p local7.info &

I invoke it as follows:

nohup startproc.sh /usr/local/bin/app

The output from the program is piped to logger, which streams it to syslog daemon. The logs go with a timestamp by default and a tag of program name. I use “local7.info” priority; there are other possibilities.

syslogd can redirect the logs to any file. I have set up my /etc/syslog.conf as follows:

*.*     /var/log/app.log

After adding the configuration, re-initialize the syslog via:

killall -HUP syslogd

This causes logs with the tag app to be written to app.log.

Note: If you do not already have an app.log, syslog will create it, as root. This means if your application runs under a different user, it cannot write to the log. Create the log file yourself before sending the HUP signal to syslogd.

Rotating the log files

newsyslog is a FreeBSD utility that can rotate logs. Linux users can run logrotate instead. I have the following configuration in /etc/newsyslog.conf:

/var/log/app.log        user:group      644     10      10000    *       GZ

newsyslog is invoked by the system automatically on an hourly basis. If you want to know how, you can look at the following /etc/crontab entry:

#minute hour    mday    month   wday    who     command
0       *       *       *       *       root    newsyslog

It’s worth repeating that you do not have to do anything to invoke newsyslog. It is done by the system automatically. You only need to set up newsyslog.conf.

Once rotation is done, you will see log files as shown earlier. If you want to test your changes by forcing a rotation, you can simply say:



The advantages of this design is that you do not need any external libraries. The program can be written in any language, but log rotation is handled in a uniform way. You can use syslog to redirect output to a log file as well as an external centralized logging server. You can continue to run the program for debugging purposes and it will print to the console.

A disadvantage of this design is that you have the configuration spread across multiple files. But we use a configuration management tool, Ansible, and all the steps are contained in a single playbook. So this is not a problem.

For additional information, see the FreeBSD handbook on configuring system logging.

See also