Thoughts on backend and scaling

Logging into Syslog with Go

Logging is so simple that it is very hard to implement it right. In a project I started to clear up logging and I noticed that there are many places one can introduce syslog as a logging target, and there are a lot of things can get just too messy. I want to show how I started with syslog, how I separated the different layers of logging.

To go on I suppose you have rsyslog already installed on your Linux.

What is syslog?

Syslog is simple and fast logging system built in the Posix operating systems. You can think that it is a logging hub, there are several inputs it can consume, it can emit logs to multiple outputs and also it has a rich rule engine.

When an application logs it can choose which syslog facilities wants to send log messages. There are some builtin facilities like KERN, USER, DAEMON, CRON and there are custom facilities for local applications from LOCAL0 to LOCAL7 (at least).

Let the application log

So at first our application should connect to those facilities.

package main

import (
	"flag"
	stdlog "log"
	"log/syslog"
)

func main() {
	logger, err := syslog.NewLogger(syslog.LOG_LOCAL0|syslog.LOG_INFO, stdlog.Lshortfile)
	if err != nil {
		panic(err)
	}

	audit, _ = syslog.NewLogger(syslog.LOG_LOCAL1|syslog.LOG_INFO, stdlog.Lshortfile)
}

That is pretty simple we made a generic logger which writes to the first local facility at info level, so we can refer to this action as local0.info later. The second log we will write rarely, it will be an audit log syslog.

It is better to use a logging framework, I am using the gommon/log framework, pretty straightforward to set up. So let us connect the logger to the log framework.

package main

import (
	"flag"
	stdlog "log"
	"log/syslog"

	"github.com/labstack/gommon/log"
)

func main() {
	logger, err := syslog.NewLogger(syslog.LOG_LOCAL0|syslog.LOG_INFO, stdlog.Lshortfile)
	if err != nil {
		panic(err)
	}

    log.DisableColor()
    log.SetOutput(logger.Writer())
    log.SetHeader("${time_rfc3339_nano} ${level} ${short_file}:${line} ${prefix}")

	audit, _ = syslog.NewLogger(syslog.LOG_LOCAL1|syslog.LOG_INFO, stdlog.Lshortfile)

    log.Info("This is an info log message")
    log.Warn("This is a warning")
}

Let us run this app, what happens? In the /var/log/messages you will see two lines

Jun 30 08:49:31 myhost /home/richard/logtest[53676]: \
  2020-06-30T08:49:31.701738949Z INFO main.go:27 - This is an info log message
Jun 30 08:49:31 myhost /home/richard/logtest[53676]: \
  2020-06-30T08:49:31.701852547Z WARN main.go:29 - This is a warning

Why?

rsyslog.conf

In default the rsyslog writes every info level message to that file. If you check the /etc/rsyslog.conf you can see this line.

*.info;mail.none;authpriv.none;cron.none    /var/log/messages

It means that *.info so local0.info also written to the messages log. You can see in the code that we also added the LOG_INFO when we made the syslog loggers.

It would be nice to separate those logs from the system messages, right? So at first we need to redirect them to a different file, and also we need to remove from the messages log. The first is a piece of cake, just add a /etc/rsyslog.d/my-app.conf with this content.

$FileCreateMode 0644
local0.*        /var/log/my-app.log
$FileCreateMode 0600

At first we set the file create mode in order that everybody can read the log file. Let us suppose that not the root user will run our app, so it would be nice if that user can read the application logs. The setting is global, so from now on this file mask will be effective. That is why we set it back to 0600.

Let us restart the rsyslog.service.

Well, we are half-ready, we need to remove the local0.* from messages. We need to change the messages rule to this

*.info;mail.none;authpriv.none;cron.none;local0.none    /var/log/messages

So from the local0 log facility nothing will be written to the messages.

Formatting log messages

We are almost done, I just don’t want to see two log message timestamps in the logs. One is put by the Go logging framework, and the other is done by syslog. I trust syslog, so in the application we can remove the timestamp generating for the log messages.

    log.SetHeader("${level} ${short_file}:${line} ${prefix}")

That is nice but the default syslog timestamp is pretty vague. We need to explain syslog to use a more precise time format. The RFC3339 is with microsecond precision, most of the cases is good enough.

$FileCreateMode 0644

$template myFormat,"%TIMESTAMP:::date-rfc3339% %HOSTNAME% %msg%\n"
local0.*		/var/log/my-app.log;myFormat

$FileCreateMode 0600

Log lines should be like this

2020-06-30T10:54:31.693715+00:00 myhost WARN main.go:29 - This is a warning

You can setup the audit log similarly, to go all the audit logs to a different file. If you don’t need the

So that is it, I needed to collect the different information fragements from different sources, I hope it will help you to setup some clear logging of your applications.