Study of a docker-compose cluster whose overall health is checked by a dedicated container; in essence, it runs smoke tests from inside the cluster network. The actual checks being run are defined in healthchecks.json.
Inspired by an actual issue at work, this mini project mainly serves as investigation grounds for which language provides the nicest way to implement "scripts" in a container context. Criteria are:
- Easy to write correct scripts. Ideally testable.
- Good abstractions for readable scripts.
- Minimal tooling required and available as Docker images.
- Low runtime requirements for the final container.
Yes, I want to get rid of the ubiquitous "small" shell script as a default approach.
More specifically, the following language or library features are needed.
- Command-line parameters.
- Environment variables.
- JSON parsing & generation.
- File I/O.
- External command execution.
First, some numbers. Smaller is better.
| LOC | LOD | SOF | SLI | |
|---|---|---|---|---|
| Go | 135 | 15 | 5.4MB | 356MB |
Where the metrics are defined as follows:
- LOC -- Lines of "script" (and tooling) code
- LOD -- Lines in Dockerfile.
- SOF -- size overhead in the final image,
that is the size of the final image minus the size of
healthcheck. - SLI -- size of the largest intermediate image.
And now some subjective observations.
- The standard library (or, for that matter, the official Docker container) has everything we need here.
- No tooling beyond the compiler required.
- Code is reasonably neat and could be tested.
- Struct-based JSON marshalling makes for safe code.
- Some gaps in the standard library that are off for a modern language, e.g. higher-order collection functions.
docker build -t healthcheck --build-arg checks=healthchecks.json healthcheck/
docker-compose -d upCheck docker-compose.yml for the port mappings of the individual healthcheckers.