Skip to content

DESIGN.md

Latest commit

 

History

History
37 lines (28 loc) · 3.52 KB

DESIGN.md

File metadata and controls

37 lines (28 loc) · 3.52 KB

Use Case

In a service oriented architecture, an existing challenge is to manage running binaries, dependencies and networking. Orchestra's goal is to provide a unique place where you can run, stop, aggregate logs and config your Go binaries.

Goals

  • Standalone: The first requirement is to provide support as a standalone tool, without external dependencies (apart from docker when using vendors). An optional flag will be provided to run the Go binaries inside a docker container.
  • Rerun: Reload (build, test and run) services upon modification
  • Vendors: Specify dependecies with existing services. Vendor software (e.g. postgres, rabbitmq, etc) will run inside Docker. This feature relies on crane.
  • Configuration: A global configuration file will be required to specify ENV variables for every service. An optional configuration file can be specified
  • Testing: Run unit tests and acceptance tests.
  • Logging: Show aggregated logs for every service in the fleet.
  • Reliability: Services started with Orchestra should operate atomically and in a reliable way. When Orchestra starts a service should check for the running processes, and match and kill services running outside this toolkit (e.g. running go run main.go inside a service folder or the ./service binary)
  • Scale: Services should be able to scale, by default the scale level is set to 1. A scale value can be set for every service inside their configuration file.

Command Line Cmd/Flags Proposal

  • start
    • rerun: uses rerun to run the services
    • scale: overwrites the defined scale parameter
  • stop
  • restart
  • vendors
  • logs
  • ps

Design Proposal

  • The application will create a .orchestra folder in the directory (inside the GOPATH) that we want to orchestrate. The purpose of this directoy is to provide the necessary context when running orchestra again.
  • The application can be run in every folder inside the GOPATH, e.g. let's assume that we have a single repo (called myservices) with all the services in different folders, you can run orchestra inside that folder and it will recursively (with depth=1) look in every directory for a configuration file (i.e. service.yml) and register the services. This approach works seamlessly with services split in multiple repositories in one GitHub account.
  • Logging is extremely important, and with great logging comes great responsibilities. For simplicity we'll redirect the output of a service into a file, then we use github.com/ActiveState/tail library to aggregate the logs of different services, maybe make them colorful and prefix them with the service name.
  • Code reload of Go binaries can be achieved using github.com/skelterjohn/rerun. It relies on fsnotify and it watches for changes in the Go files. It can be possible, before every run, to run the tests again using go test. I am considering of forking the project for the sake of better logging.
  • The application is a pure command-line toolkit, for this reason we may need commands, maybe subcommands and flags. For simplicity I intend to use the well known github.com/codegangsta/cli library.
  • YAML will be used as the standard format to store configuration files. In Go, a YAML parser is required, github.com/go-yaml/yaml is my favorite option at the moment to marshal or unmarshal files.
  • An initialization process may be used to initialize a folder as an orchestra project. This may be a useful step to do some checks before start looking into all subdirectories and start running the binaries.