Skip to content

Latest commit

 

History

History
95 lines (66 loc) · 4.43 KB

ARCHITECTURE.md

File metadata and controls

95 lines (66 loc) · 4.43 KB
Raw

Architecture

The idea behind OTel recipes is: Wouldn't it be nice if there was a list of steps that I could follow and, in the end, I "have OTel in my app"?

This is what guides the entire architecture and design decisions of the project. In its core, OTel recipes MUST

  • Be easy and approachable for newcomers in OpenTelemetry
  • Teach people "how do I do X" with OpenTelemetry
  • Be certain that if the steps are followed it will work. There's nothing more frustrating than copying and pasting some snippets only to realize they don't work.

With this premise, let's dive deep into the components that comprise OTel recipes.

Recipe file (recipefile.json)

At the core, there's the recipe file JSON schema. The recipefile.json can be understood as the metadata about each recipe application. The recipe file explains what the sample application is about, which telemetry signal (traces, metrics, logs) it is teaching users, and, more importantly, which steps the user must do to achieve the goal.

The idea behind the steps property in the recipe file is to show exactly what/where users need to change in their applications. For example, if the goal is: Configure my application to emit OTel traces, a list of steps would be:

  1. Install the dependencies
  2. Configure the OTel SDK for traces
  3. Record a span with an attribute

Each step then links to the actual code file of the recipe in the repository via the source property. This property is the most important and it is what is used to rendered the code step in the website!

The recipe files are also checked via GitHub workflows against the JSON schema. So, if something is wrong it will be caught during CI checks 😉. (See Check recipe file workflow).

Sample applications

The sample applications are the bulk of content in the project. To help searching/filtering, the samples are categorized in programming language and telemetry signal.

src/
└── csharp/
    └── traces/
    └── metrics/
    └── logs/

Along with its recipefile.json, each application MUST declare its Dockerfile, a Go test module and a docker-compose.yaml. During CI, the application is started and its declaring integration test is executed.

This all happens in the workflow Build and Test:

build architecture

Thanks to the file conventions, the GitHub workflow becomes simple and generic: At the same hierarchy a recipefile.json is found, we can also assume there's a docker-compose.yml file that knows how to start everything it is needed to run and test the sample application.

Another benefit of the structure above is: each sample application can define and use its own dependencies without "poluting" the rest of samples. For example, if a sample wants to demonstrate how to instrument async scenarios, e.g., via messaging systems, it can add the messasing system as a dependency in the compose file. Nothing else needs to change 😎.

Tests

To learn how to use the test framework/utilities, see Test utils.

What is left for the actual integration test is:

  • Fetch the app's telemetry
  • Run assertions

That is when the goal of keeping the use-cases for the sample apps simple and minimal comes in handy. With the test utils, we can query the OTLP back-end and "find" the spans/metrics/logs we are looking for.

To learn how to add a a integration test for a recipe application, see Testing a recipe

Website

To tied things together we have the website - a single page application built with Sveltekit.

The website runs completely on the client side. There's no back-end service the website's communicate with to get the recipes data. So how does it work, you ask?

The website uses the recipefile.json files from all the apps!. During a release, all the recipefile.json are bundled together (using jq) into a single .json file which is then minified and embedded in the website.

For starting, this approach will be probably enough, apart from making the website super snappy 🚀 as there's nothing to be loaded.

Then the website simply rely on the recipefile.json structure to display all the information about the recipe application, specially the individual steps on how to get it configured with OpenTelemetry.