Tracing is used to reproduce specific scenarios and request flows. It's already helpful in retracing complex application processes, but it's especially helpful when multiple applications and instances are involved.

However, what's important to be pointed out is that there needs to be a business, not technical, requirement for tracing systems, similar to journaling.

Tracing is a poor technique for debugging or performance tracing systems. It will have a certain impact on performance and doesn't help much in resolving performance regression. Interdependent, distributed applications that need to be optimized in their performance advisably solely emit information about their quality of service, such as response times. Sampling techniques can sufficiently gather information that indicate performance issues in the applications.

However, let's have a look at business-motivated tracing to track the components and systems involved.

The following diagram shows a trace of a specific request involving multiple application instances and components thereof:

The trace can also be displayed in a timeline to show the synchronous invocations as demonstrated in the following diagram:

Tracing includes information about which applications or application components have been involved and how long the individual invocations took.

Traditionally, log files have been used for this, by logging the start and end of each method or component invocation including a correlation ID, such as a thread identifier. There is the possibility of including correlation IDs into logs that are used from a single originating request and are reused and logged in subsequent applications. This results in traces that also span multiple applications.

In the case of logging, the tracing information was accumulated from multiple log files; for example, using solutions such as the ELK stack. Trace logs are usually implemented in cross-cutting ways; for example, using logging filters and interceptors so as not to obfuscate the code.

However, using log files for tracing is not advisable. Even enterprise applications that experience a moderate load introduce a lot of log entries that are written to files. Many log entries are needed for each and every request.

File-based I/O and the needed log format serialization generally is too heavy for this approach and greatly affects the performance. Tracing to log file formats introduces a lot of data that needs to be parsed again afterwards.

There are tracing solutions that provide a much better fit.