Skip to content

palantir/go-baseapp

Repository files navigation

go-baseapp GoDoc

A minimal, not-quite-a-framework for building web applications on top of the standard library. It provides:

  • A selection of dependencies for logging, metrics, and routing that fit well with the standard library
  • A standard configuration type
  • A basic configurable server type
  • A default (but optional) middleware stack

This doesn't take the place of frameworks like echo, gin, or others, but if you prefer to stay close to the standard library for simple applications, go-baseapp will save you time when starting new projects.

Usage

Create a baseapp.Server object, register your handlers, and start the server:

func main() {
    config := baseapp.HTTPConfig{
        Address: "127.0.0.1",
        Port:    8000,
    }
    loggingConfig := baseapp.LoggingConfig{
        Pretty: true,
        Level: "debug",
    }

    logger := baseapp.NewLogger(loggingConfig)

    // create a server with default options and no metrics prefix
    server, err := baseapp.NewServer(config, baseapp.DefaultParams(logger, "")...)
    if err != nil {
        panic(err)
    }

    // register handlers
    server.Mux().Handle(pat.Get("/hello"), http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
        baseapp.WriteJSON(w, http.StatusOK, map[string]string{
            "message": fmt.Sprintf("Hello, %s!", r.FormValue("name")),
        })
    }))

    // start the server (blocking)
    server.Start()
}

Example

The example package provides a full server that responds to GET /api/message requests with a static message from the configuration.

You can start the server using:

./godelw dep
./godelw run example

Navigate to http://localhost:8000/api/message to see the output.

Features and Opinions

go-baseapp was designed and is maintained by Palantir's developer tools team. When developing this project, we had the following in mind:

  1. Minimize code and knowledge beyond the standard library. Because we support many applications that have many contributors, we want to minimize the time people spend learning frameworks before being able to add new features or develop new services. To this end, we've used the standard library where we can and picked small dependencies with obvious APIs when we need more features.

  2. Allow easy extension. Our opinions in this library form a starting point, but it should be possible to ignore any of them and add your solutions if you disagree with something or it doesn't work in a specific situation.

  3. Provide ops-friendly defaults. Our default configuration enables things like request IDs, request and runtime metrics, nice error formatting, and panic recovery. These are all easy to implement, but are also easy to forget when setting up a new project.

Notably, performance is not a major concern. While we try to avoid decisions that are obviously not performant, our request scale is usually low and it makes more sense to optimize on the factors above.

Dependencies

Our selected dependencies are ultimately somewhat arbitrary, but have worked well for us and do not include transitive dependencies of their own:

  • rs/zerolog for logging. We like that it has an easy-to-use API, built-in support for context.Context, and helpful net/http integration via the hlog package.
  • rcrowley/go-metrics for metrics. We like that it supports many metrics types, isn't coupled to a specific publish method, and has existing integrations with many monitoring and aggregation tools.
  • goji.io for routing. We like that it provides a minimal API, supports middleware, and integrates nicely with context.Context.
  • bluekeyes/hatpear for error aggregation. We like that it allows returning errors from HTTP handlers in a way that integrates well with net/http.

Server Options

The default server configuration (baseapp.DefaultParams) does the following:

  • Sets a logger
  • Creates a metrics registry with an optional prefix
  • Adds the default middleware to all routes (see below)
  • Configures log timestamps to use UTC and nanosecond precision
  • Improves the formatting of logged errors
  • Enables Go runtime metrics collection

If you only want some of these options, provide your own set of parameters with only the parts you want; all of the components are exported parts of this library or dependencies.

Graceful Shutdown

go-baseapp can be optionally configured to gracefully stop the running server by handling SIGINT and SIGTERM.

The parameter ShutdownWaitTime on the baseapp.HTTPConfig struct enables graceful shutdown, and also informs the server how long to wait during the shutdown process before terminating.

Middleware

The default middleware stack (baseapp.DefaultMiddleware) does the following:

  • Adds a logger to the request context
  • Adds a metrics registry to the request context
  • Generates an ID for all requests and sets the X-Request-ID header
  • Logs and emits metrics for all requests
  • Handles errors returned by individual route handlers
  • Recovers from panics in individual route handlers

If you only want some of these features, create your own middleware stack selecting the parts you want; all of the components are exported parts of this library or dependencies.

Metrics

If enabled, the server emits the following metrics:

metric name type definition
server.requests counter the count of requests handled by the server
server.requests.2xx counter like server.requests, but only counting 2XX status codes
server.requests.3xx counter like server.requests, but only counting 3XX status codes
server.requests.4xx counter like server.requests, but only counting 4XX status codes
server.requests.5xx counter like server.requests, but only counting 5XX status codes
server.requests.latency timer the latency of requests handled by the server
server.requests.2xx.latency timer like server.requests.latency, but only counting 2XX status codes
server.requests.3xx.latency timer like server.requests.latency, but only counting 3XX status codes
server.requests.4xx.latency timer like server.requests.latency, but only counting 4XX status codes
server.requests.5xx.latency timer like server.requests.latency, but only counting 5XX status codes
server.goroutines gauge the number of active goroutines
server.mem.used gauge the amount of memory used by the process in bytes

Note that the server.requests.latency timers may be used instead of the basic server.requests counters, as they include counts and rates in addition to the latency distribution.

The appmetrics/emitter/datadog package provides an easy way to publish metrics to Datadog. Timer metrics are reported in nanoseconds. When exporting these metrics from the registry, you may wish to convert the units, for instance by calling datadog.SetTimerUnit(time.Millisecond).

The appmetrics/emitter/prometheus package provids an easy way to expose metrics on a Prometheus-compatible endpoint.

Contributing

Contributions and issues are welcome. For new features or large contributions, we prefer discussing the proposed change on a GitHub issue prior to a PR.

New functionality should avoid adding new dependencies if possible and should be broadly useful. Feature requests that are specific to certain uses will likely be declined unless they can be redesigned to be generic or optional.

Before submitting a pull request, please run tests and style checks:

./godelw verify

License

This library is made available under the Apache 2.0 License.