Set log level from a new LOG_LEVEL env var.

[athamark]

During live debug sessions we realized that the AuthService logs are way too verbose and thus not helpful for debugging. Since all requests are passed through AuthService for authentication we have a chunk of logs for every one of the requests, even for requests to whitelisted URLs.

Related upstream PR

As you can see in the repo there is an upstream PR which dates back to 2019 (#3). This PR introduces a new LOG_LEVEL envvar that expects one of the following values:

• warn
• info
• debug

The functionality that the contributor has added seems to be in the right direction. But a lot has changed since they submitted this PR. So we had better start fresh and lay out a plan that complies with the latest oidc-authservice.

Proposed Plan

We need to re-categorize our log messages in the available log level categories. Right now the log level which we use in oidc-authservice are:

• fatal
• error
• warn
• info
• debug

The current log messages that belong to either one of the first three log levels should remain in these log levels. The same goes for the log messages that belong to the debug log level.

The challenge is to revise the log messages of the info log level which is currently way too verbose and leads to unreadable logs. The plan is to distinguish which of the info log messages should belong to the debug log level instead. As we will expose later on, we will also transfer a few settings-related info logs to the warn log level.

Good to have

It would be beneficial to isolate the dependency to the "github.com/siroupsen/logrus" (https://github.com/sirupsen/logrus) package inside the common package. For this we will need a new and simple function that sets up a StandardLogger():

func StandardLogger() *log.Logger {
	return log.StandardLogger()
}

This will make our code more maintainable and also it will allow us to enforce a single log level across our code.

Regarding the order of the log levels

As exposed in the README.md of the "github.com/sirupsen/logrus" library (https://github.com/sirupsen/logrus):

You can set the logging level on a Logger, then it will only log entries with that severity or anything above it:
// Will log anything that is info or above (warn, error, fatal, panic). Default.

   log.SetLevel(log.InfoLevel)

It may be useful to set log.Level = logrus.DebugLevel in a debug or verbose environment if your application has that.

From this I deduce that the ordering is:

1. panic (we have no such log message in AuthService)
2. fatal
3. error
4. warn
5. info
6. debug

Setting the log level to the i-th item means that all the log messages that belong to the {1, .., i} set will be logged. If a log message belongs to {i+1, ..., 6} then it will not be logged. So the bigger the configured log level is, the more verbose the logs of AuthService will be.

Let's default to info log level.

Testing

I made a rather simple test to quantify the impact of these changes in the AuthService logs. The 5-minute experiment I did was the following:

1. start AuthService pod (0:00)
2. login as [email protected] (2:00)
3. keep the UI as an active browser window
4. logout (4:00)
5. stop the AuthService logs

I did this for both an old and the new image. The comparison is:

• old image logged 452 lines
• new image logged 70 lines.

Thanks for your time!