go-gracefully

go-gracefully is a library for monitoring and reporting on an applications' health. Unlike many solutions out there, go-gracefully uses an asynchronous, stream based check scheme. This enables real-time checks as well as push based changes.

Status:

This library is currently available as a preview. I started development to support my existing work on deps.cloud.

Most health check libraries you find are pull based. In this model, you define your check as a function that's called on some set interval. While this is a great start to a solution, more advance techniques need to be able to push.

The flow of information in the system is as follows:

1. Each Check produces a Report
2. Reports are collected through the use of a channel by the Monitor.
3. The monitor updates its summary of the system.
4. When either a check, or the system state changes, a Report is published to subscribers.

A snapshot of the full report can be obtained from the Monitor.

How does it work?

All checks in go-gracefully have a common block of metadata.

• Name is a required descriptor of the check.
  • Avoid whitespace where possible. Preferred character set: [a-zA-Z0-9-_.]
• An optional Runbook can be provided to aid in the resolution of issues.
  • When provided, this should be a valid URL.
• Finally, a Weight is used to determine relative importance of check to the overall system.
  • A check with Weight: 100 has a greater impact on the system health than one with Weight: 10.

    // ...
    Metadata: check.Metadata{
        Name: "periodic-check",
        Runbook: "http://path/to/runbook.md",
        Weight: 10,
    },
    // ...

When check's evaluated, it can return one of four possible states.

• OK - The check is operating as expected.
• Minor - The check is failing and will require attention soon.
• Major - The check is failing and requires attention soon.
• Outage - The check is failing and requires attention.

On its own, state represents a fractional value of health (i.e. [0-1]). Together, the state and the weight are used to approximate an applications' health.

Inspirations

There are a lot of prior work out there. It's hard to list them all. The list below was is just a few I drew inspiration from.

• http://github.com/indeedeng/status
• https://github.com/InVisionApp/go-health
• https://godoc.org/github.com/heptiolabs/healthcheck
• https://github.com/AppsFlyer/go-sundheit

Installation

go get -u github.com/mjpitz/go-gracefully

Usage

package main

import (
    "context"
    "http"
    "log"
    "time"

    "github.com/mjpitz/go-gracefully/check"
    "github.com/mjpitz/go-gracefully/health"
    "github.com/mjpitz/go-gracefully/state"
)

func main() {
    monitor := health.NewMonitor([]check.Check{
        &check.Periodic{
            Metadata: check.Metadata{
                Name: "periodic-check",
                Runbook: "http://path/to/runbook.md",
                Weight: 10,
            },
            Interval: time.Second * 5,
            Timeout: time.Second,
            RunFunc: func(ctx context.Context) (state.State, error) {
                // make API call, check system health
                return state.OK, nil
            },
        },
        &check.Stream{
            Metadata: check.Metadata{
                Name: "stream-check",
                Runbook: "http://path/to/runbook.md",
                Weight: 10,
            },
            WatchFunc: func(ctx context.Context, channel chan check.Result) {
                stopCh := ctx.Done()

                var upstreamCh chan interface{}
                // make call that fills chan

                for {
                    select {
                        case <-stopCh:
                            return
                        case _ = <-upstreamCh:
                            channel <- check.Result{
                                State: state.OK,
                                Error: nil,
                            }
                    }
                }
            },
        },
    }...)

    reports, unsubscribe := monitor.Subscribe()
    defer unsubscribe()

    ctx := context.Background()
    if err := monitor.Start(ctx); err != nil {
        log.Fatal(err.Error())
    }

    // subscribe to changes in health
    for report := range reports {
        // access check information if present
        // - Check will not be present for changes in overall system health
        _ = report.Check

        // access check evaluation result data
        // - Result will be present for all reports
        _ = report.Result
    }

    // or add an HTTP endpoint to view the results of it
    http.HandleFunc("/healthz", health.HandlerFunc(monitor))
}