The bork application is a sample application to demonstrate Go API patterns and provide a lightweight template for a new application.
Rather than attempting to create a monolithic CRUD/MVP framework based app, this application centers a set of "services" which contain business logic for the application. API and data layers are then orchestrated to serve and retrieve data defined by the services.
Note, this application is mostly a guide for developing applications with a particular set of patterns. Adopting it requires some orchestrating based on your business context (ex. plugging in an auth solution). It attempts to illustrate concepts that have been repeated in multiple Truss environments, while implying space for domain specific changes. Some basic uses of this repo are:
- Forking some boilerplate code commonly written at application bootstrap.
- Looking for Go API patterns to employ.
- Adding sample API or data sources. Want to show what a GraphQL server looks like as opposed to standard library handlers? APIs would be a great spot. Maybe there's a new sql adapter-- take a look at Data Sources.
- Testing out a new pattern on a minimal dependency application. While a production application may have compiled layers or history and third party dependencies, this API mostly requires Go and Postgres and has minimal "business" constraints.
The main patterns of the API are:
- Service package for business logic. Business logic is scoped to a single package using single functions per instruction of work. Since application development is led by client user value, the goal is to focus on adding value to this layer of the code. Feature oriented work will tend to be mirrored in this layer.
- Composable API and data source layers. The API and data source layers are treated as tools to serve the logic of the application. Rather than overloading those layers with logic (such as monolithic handlers), they serve small, scoped sets of responsibilities. Separating these packages allows the tooling of the application to progress without affecting the business logic (ex. migrating from REST to GraphQL). A goal is to spend less time in these layers and automate out of as much of the boilerplate (ex. generating code) to allow for focus on the service layers.
- Cross package domain models. Few resources are shared across layers of the application and these resources are almost entirely scoped to method-less data structures and Go basic types or standard library definitions. These include: models, errors, and context. Reducing the function of these types heavily reduces coupling of the application's components.
- Internally defined, minimal, dependency/parameter interfaces. When defining dependencies for inter-layer communication, interfaces are defined internally to the calling package. These interfaces are scoped to fewest methods necessary and often defined as typed functions. These implementations of interfaces are set up in a main server package and explicitly defined. Elsewhere in the application, functions are almost exclusively referred to as interfaces or typed functions. Paired with minimal domain models, this reduces layer coupling and allows implementation packages to implicitly satisfy caller requirements.
Note that many of these patterns pull from SOLID principles and attempt to implement them in line with Go's language design.
The application also consists of a few dependable libraries used in Truss codebases, such as:
- Testify for testing suites and sub-test structuring.
- sqlx for marshaling sql into Go structs.
- mux mostly used for simple sub-routing.
- viper for app configuration.
- clock It's a clock interface and mock clock.
Mostly used for
time.Now()
mocking in tests. - zap A structured logger with nice syntax for adding fields.
For more information on each package, see codebase layout.
cmd
is the entry-point
for running the application from the command line.
It's lightweight and holds very few responsibilities.
Namely:
- Read the command line arguments.
- Initialize environment variables.
- Execute the application.
In a production application, the may also hold some helper scripts for running maintenance or testing tasks on the application.
pkg
holds the necessary code
for serving the application.
This is distinct from cmd
in that is should be executable from multiple environments
including, cmd
, integration testing or other Go code.
In relation to cmd
, it takes viper.Viper
as it's only argument
(in the server package),
and has no awareness of the command line or executing environment.
There are a number of things you'll need at a minimum to be able to check out, develop, and run this project.
- Install Homebrew
- Use the following command
/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
- Use the following command
- Install Go with Homebrew:
brew install go
* Note: If you have previously modified your PATH to point to a specific version of go, make sure to remove that. This would be either in your.bash_profile
or.bashrc
, and might look something likePATH=$PATH:/usr/local/opt/[email protected]/bin
. - Ensure you are using the latest version of bash for this project:
-
Install it with Homebrew:
brew install bash
-
Update list of shells that users can choose from:
[[ $(cat /etc/shells | grep /usr/local/bin/bash) ]] \ || echo "/usr/local/bin/bash" | sudo tee -a /etc/shells
-
If you are using bash as your shell (and not zsh, fish, etc) and want to use the latest shell as well, then change it (optional):
chsh -s /usr/local/bin/bash
-
Ensure that
/usr/local/bin
comes before/bin
on your$PATH
by runningecho $PATH
. Modify your path by editing~/.bashrc
or~/.bash_profile
and changing thePATH
. Then source your profile withsource ~/.bashrc
or~/.bash_profile
to ensure that your terminal has it.
-
- Note:
If you have previously used Golang, please make sure none
of them are pinned to an old version by running
brew list --pinned
. If they are pinned, please runbrew unpin <formula>
. You can upgrade these formulas instead of installing by runningbrew upgrade <formula
.
Use your work email when making commits to our repositories. The simplest path to correctness is setting global config:
git config --global user.email "[email protected]"
git config --global user.name "Trusty Trussel"
If you drop the --global
flag,
these settings will only apply to the current repo.
If you ever re-clone that repo or clone another repo,
you will need to remember to set the local config again.
You won't.
Use the global config. :-)
For web-based Git operations, GitHub will use your primary email unless you choose "Keep my email address private". If you don't want to set your work address as primary, please turn on the privacy setting.
Note that with 2-factor-authentication enabled, in order to push local code to GitHub through HTTPS, you need to create a personal access token and use that as your password.
You can checkout this repository by running
git clone [email protected]:trussworks/go-sample.git
.
Please check out the code in a directory like
~/Projects/go-sample
and NOT in your $GOPATH
. As an example:
mkdir -p ~/Projects
git clone [email protected]:trussworks/go-sample.git
cd go-sample
You will then find the code at ~/Projects/go-sample
.
You can check the code out anywhere EXCEPT inside your $GOPATH
.
So this is customization that is up to you.
- Install direnv:
brew install direnv
- Set environment with:
direnv allow
- Install pre-commit:
brew install pre-commit
- Run
pre-commit install
to install a pre-commit hook into./git/hooks/pre-commit
. - Next install the pre-commit hook libraries
with
pre-commit install-hooks
.
To build the cli application in your local filesystem:
go build -a -o bin/bork ./cmd/bork
You can then access the tool with the bork
command.
Alternatively to manually building the app, you can run the server and reload files with air:
# outside of a Go project,
# to install globally
go get -u github.com/cosmtrek/air
air
To use the database, run:
docker run --name bork-postgres \
--publish $PGPORT:5432 \
-e POSTGRES_DB=$PGDATABASE \
-e POSTGRES_PASSWORD=$PGPASS \
-d postgres
Install flyway to run migrations:
brew install flyway
Then run them with:
flway migrate
To add a new migration, add a new file to the migrations
directory
following the standard
V_${last_migration_version + 1}_your_migration_name_here.sql
Run tests with:
go test ./pkg/...
The APIs reside at localhost:8080
when running.
To run a test request,
you can send a GET to the health check endpoint:
curl localhost:8080/api/v1/healthcheck
Logging in this project is intentionally restricted to a single structured log line per request. The log line will be either Info or Error Level and will have a variety of key/value pairs associated with it. The primary reason for doing things this way is to aid in debugging issues in production by pre-bucketing all info by request. This allows you to use line-oriented tools to quickly determine if there are commonalities between errors.
The line logging is performed by the middleware in pkg/server/httpserver/logger.go, in code you interacti with it with these methods from appcontext.
//LogRequestField adds a zap.Field to the line logged at the end of the request LogRequestField(ctx context.Context, field zap.Field)
You can call this anytime you have a piece of information that you might like associated with this request. e.g.:
- user_id
- auth_type
- number_of_dogs
// LogRequestError adds a message to the request log line and also sets it to log at the Error level LogRequestError(ctx context.Context, message string, err error)
You should call this at most once in a request if you want the request to log at the Error level. The message will be logged with an "error_message" key and the error will be logged with the zap standard "error" key. In general, these should generally correspond with requests that have 5xx status codes. Error log levels may trigger notificactions and require investigation.