Documentation:

Logging

Things may go wrong. It’s computers, after all. If a log exists, developers and/or users have a chance of figuring out what happened, and fix whatever the problem was. Without such logs, life can be very frustrating. So, let’s log!

Analyzing what happened on a user’s device is easier if all code on that device uses single logging facility. If a problem occurs due to an unexpected interaction between two different components, for example, a single log has all events in chronological sequence and debugging is much easier.

Log levels

By default, all packages, Apps and Accessories should log warnings, errors and above. They should not log informational, debug-level or other low-level messages.

It should be documented how it is possible to temporarily change the level of logging, e.g. to assist the user with diagnosing and resolving a problem.

Journald-based logging

UBOS uses the logging facilities provided by systemd‘s journald. All packages, Apps and Accessories are strongly encouraged to use them.

You can find a few excellent introductions to logging via systemd on-line, e.g.

For general information about systemd, see the Systemd home page.

When logging, developers are encouraged to provide as much contextual information as needed to understanding the context of the message, in particular the AppConfigId of the installation that the log message refers to. The syslog identifier should be set to the name of the reporting package.

If Journald-based logging is not possible

Sometimes it is not possible to use systemd‘s journald. For example, a web server might log every request, and that would overwhelm the journal.

In this case, an App or Accessory should:

  • write log files into directory /ubos/log/<dir>, where <dir> is the name of the package, plus hyphen, plus AppConfigId.

    For example, if the name of the App is myapp and it has been deployed to AppConfiguration a1234567890, it should write its log files into directory /ubos/log/myapp-a123456789.

  • set up a logrotate policy that prevents those log files from growing endlessly. This is as simple as writing a file to /etc/logrotate.d.

  • cleanly close log files upon a log rotation event.

In case of a package that logs but that is not uniquely associated with an AppConfigId, it should:

  • write log files into directory /ubos/log/<dir>, where <dir> is the name of the package. For example, if the name of the package is mydbserver, it should write its log files into directory /ubos/log/mydbserver.
  • it also needs to set up a logrotate policy and cleanly close log files upon rotation.

Here is an example for the Apache web server’s logrotate policy, located in /etc/logrotate.d/httpd:

/var/log/httpd/*log {
    missingok
    sharedscripts
    compress
    postrotate
        /usr/bin/systemctl reload httpd.service 2>/dev/null || true
    endscript
}

You can see that in this configuration, log files will be compressed to save space, and after a rotation has been performed, the web server is cleanly restarted.