Skip to content

Greentube/monitoring

Repository files navigation

Greentube.Monitoring Build Status Build status codecov

Greentube.Monitoring is a library that simplifies monitoring dependencies that affect your application's availability.

It allows you to add one or more resource monitors and verify with a single call if any critical resource is down, which in turn reports your application as unavailable.

With a few lines of code you get a health check endpoint to reply to a load balancer or a monitoring tool that your application is ready to take work load or not.

Monitors can be simple delegates or one of the supported:

  • Redis
  • MongoDB
  • HTTP
  • SQL Server
  • ActiveMQ

Example for monitoring both Redis and MongoDB:

public void ConfigureServices(IServiceCollection services)
{
    services.AddMonitoring(options => {
        options.AddRedisMonitor();
        options.AddMongoDbMonitor();
    });
}

Or your own resource monitor:

public void ConfigureServices(IServiceCollection services)
{
    services.AddMonitoring(provider => {
        var dep = provider.GetRequiredService<IMyServiceMonitor>();
        dep.StartMonitor();
    });
}

With the ASP.NET Core Health Check middleware, the state can be reported like:

public void Configure(IApplicationBuilder app){
    app.UseHealthEndpoint()
}

Verifying the state of your service:

$ curl -D - http://localhost:5000/health
HTTP/1.1 200 OK
Date: Mon, 12 Jun 2017 13:37:35 GMT
Content-Length: 11
Content-Type: application/json
Server: Kestrel

{"Up":true}

Or when a critical resource is down:

$ curl -D - http://localhost:5000/health
HTTP/1.1 503 Service Unavailable
Date: Mon, 12 Jun 2017 13:40:42 GMT
Content-Length: 12
Content-Type: application/json
Server: Kestrel

{"Up":false}

Each of those monitors is a separate NuGet package (pay for play). If you don't want use the ASP.NET Core middleware but want to use MVC to report the state in a different way:

[Route("health")]
public class HealthController : Controller
{
    private readonly IResourceStateCollector _collector;

    public HealthController(IResourceStateCollector collector)
    {
        _collector = collector;
    }

    [HttpGet]
    public IEnumerable<object> AllResourceStates()
    {
        return from rm in _collector.GetStates()
               select new
               {
                   rm.ResourceMonitor.ResourceName,
                   rm.IsUp
               };
    }

    [HttpGet]
    [Route("non-critical/name")]
    public IEnumerable<string> NonCriticalResourcesName()
    {
        return from rm in _collector.GetStates()
               where !rm.ResourceMonitor.IsCritical
               select rm.ResourceMonitor.ResourceName;
    }
}

Usually you would like to provide the possibility of application graceful shutdown

That means that application should stop receiving requests while trying to complete running requests.

For that you need to tell load balancer to take out the node

You need to call Shutdown method on ILoadBalancerStatusProvider

Typical code can look like that (example is from hosting ASP.NET MVC Core application in Windows service)

    public class GSSService : WebHostService
    {
        private readonly INrgsLogger _logger;
        private readonly ILoadBalancerStatusProvider _loadBalancerStatusProvider;

        public GSSService(IWebHost host) : base(host)
        {
            _logger = host.Services.GetRequiredService<INrgsLogger>();
            _loadBalancerStatusProvider = host.Services.GetRequiredService<Startup.ILoadBalancerStatusProvider>();
        }

        protected override void OnStopping()
        {
            _loadBalancerStatusProvider.Shutdown 
            Thread.Sleep(30000); // waiting for LB to pickup the change

            _logger.Information(LogCategory.ApplicationLifecycleLog, "OnStopping method called.");
            base.OnStopping();
        }
    }

For a logging, standard ILogger is used implicitly via DI, so for example HttpMonitoringLogger will use ILogger.

Mostly used log level is Information (or Trace for debug), but for *Monitor it's a special case: if *Monitor is Down and it is critical then level is ERROR, for non-critical is WARNING.

License

Licensed under MIT