-
Notifications
You must be signed in to change notification settings - Fork 10.4k
Using HealthChecks
ASP.NET Core 5.0 HealthChecks package is used in all APIs and applications of eShopOnContainers.
All applications and APIs expose two endpoints (/liveness
and /hc
) to check the current application and all their dependencies. The liveness
endpoint is intended to be used as a liveness probe in Kubernetes and the hc
is intended to be used as a readiness probe in Kubernetes.
Here is the documentation about how implement HealthChecks in ASP.NET Core 5.0:
Also, there's a nice blog post on HealthChecks by @scottsauber https://scottsauber.com/2017/05/22/using-the-microsoft-aspnetcore-healthchecks-package/
The readiness endpoint (/hc
) checks all the dependencies of the API. Let's take the MVC client as an example. This client depends on:
- Web purchasing BFF
- Web marketing BFF
- Identity API
So, following code is added to ConfigureServices
in Startup
:
services.AddHealthChecks()
.AddCheck("self", () => HealthCheckResult.Healthy())
.AddUrlGroup(new Uri(configuration["PurchaseUrlHC"]), name: "purchaseapigw-check", tags: new string[] { "purchaseapigw" })
.AddUrlGroup(new Uri(configuration["MarketingUrlHC"]), name: "marketingapigw-check", tags: new string[] { "marketingapigw" })
.AddUrlGroup(new Uri(configuration["IdentityUrlHC"]), name: "identityapi-check", tags: new string[] { "identityapi" });
return services;
Four checkers are added: one named "self" that will return always OK, and three that will check the dependent services. Next step is to add the two endpoints (/liveness
and /hc
). Note that the /liveness
must return always an HTTP 200 (if liveness endpoint can be reached that means that the MVC web is in healthy state (although it may not be usable if some dependent service is not healthy).
app.UseHealthChecks("/liveness", new HealthCheckOptions
{
Predicate = r => r.Name.Contains("self")
});
The predicate defines which checkers are executed. In this case for the /liveness
endpoint we only want to run the checker named "self" (the one that returns always OK).
Next step is to define the /hc
endpoint:
app.UseHealthChecks("/hc", new HealthCheckOptions()
{
Predicate = _ => true,
ResponseWriter = UIResponseWriter.WriteHealthCheckUIResponse
});
In this case we want to run all checkers defined (so, the predicate will always return true to select all checkers).
Helm charts already configure the needed probes in Kubernetes using the healthchecks, but you can override the configuration provided by editing the file /k8s/helm/<chart-folder>/values.yaml
. You'll see a code like that:
probes:
liveness:
path: /liveness
initialDelaySeconds: 10
periodSeconds: 15
port: 80
readiness:
path: /hc
timeoutSeconds: 5
initialDelaySeconds: 90
periodSeconds: 60
port: 80
You can remove a probe if you want or update its configuration. Default configuration is the same for all charts:
- 10 seconds before k8s starts to test the liveness probe
- 1 sec of timeout for liveness probe (not configurable)
- 15 sec between liveness probes calls
- 90 seconds before k8s starts to test the readiness probe
- 5 sec of timeout for readiness probe
- 60 sec between readiness probes calls
- System requirements
- Development setup
- Databases & containers
- Architecture
- Application
- Code
- Logging and Monitoring
- Tests