Skip to content

Aggregate healthcheck that is currently used for Kubernetes cluster

Notifications You must be signed in to change notification settings

Financial-Times/upp-aggregate-healthcheck

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

upp-aggregate-healthcheck

Circle CI Go Report Card Coverage Status

The purpose of this service is to aggregate the healthchecks from services and pods in the Kubernetes cluster.

Introduction

In this section, the aggregate-healthcheck functionalities are described.

Get services health

A service is considered to be healthy if it has all the pods healthy. To determine which pods are healthy, Aggregate Healthcheck service checks each pod's __health endpoint.

Note that for services are grouped into categories, therefore there is the possibility to query the aggregate-healthcheck only for a certain list of categories. If no category is provided, the health status of all services will be displayed.

Get pods health for a service

The healths of the pods are evaluated by querying the __health endpoint of apps inside the pods. Given a pod, if there is at least one check that fails, the pod health will be considered warning or critical, based on the severity level of the check that fails.

Acknowledge a service

When a service is unhealthy, there is a possibility to acknowledge the warning. By acknowledging all the services that are unhealthy, the general status of the aggregate-healthcheck will become healthy (it will also mention that there are 'n' services acknowledged).

Sticky categories

Categories can be sticky, meaning that if one of the services become unhealthy, the category will be disabled, meaning that it will be unhealthy, until manual re-enabling it. There is an endpoint for enabling a category.

Running locally

To run the service locally, you will need to run the following commands first to get the vendored dependencies for this project:

go build

There is a limited number of functionality that can be used locally, because we are querying all the apps, inside the pods and there is no current solution of accessing them outside of the cluster, without using port-forwarding. The list of all functionality that can be used outside of the cluster are:

  • Add/Remove acknowledge
  • Enable/Disable sticky categories

Build and deployment

To build Docker images for this service, use the following repo: coco/upp-aggregate-healthcheck

How to configure services for aggregate-healthcheck

For a service to be taken into consideration by aggregate-healthcheck it needs to have the following:

  • The Kubernetes service should have hasHealthcheck: "true" label.
  • The container should have Kubernetes readinessProbe configured to check the __gtg endpoint of the app
  • The app should have __gtg and __health endpoints.

How to configure categories for aggregate-healthcheck

Categories are stored in Kubernetes ConfigMaps. The template of a ConfigMap for a category is shown below:

  kind: ConfigMap
      apiVersion: v1
      metadata:
        name: category.CATEGORY-NAME # name of the category
        labels:
          healthcheck-categories-for: aggregate-healthcheck # this flag is used by aggregate-healthcheck service to pick up only ConfigMaps that store categories.
      data:
        category.name: CATEGORY-NAME # name of the category
        category.services: serviceName1, serviceName2, serviceName3 # services that belong to this category
        category.refreshrate: "60" # refresh rate in seconds for cache (by default it is 60)
        category.issticky: "false" # boolean flag that marks category as sticky. By default this flag is set to false.
        category.enabled: "true" # boolean flag that marks category as disabled. By default, this flag is set to true.

Endpoints

In the following section, aggregate-healthcheck endpoints are described. Note that this app has two options of retrieving healthchecks:

  • JSON format - to get the results in JSON format, provide the "Accept: application/json" header
  • HTML format - this is the default format of displaying healthchecks.

Service endpoints

Note that there is a configurable pathPrefix which will be the prefix of each endpoint's path (E.g. if the prefix is __health, the endpoint path for add-ack is __health/add-ack. The default value for pathPrefix is the empty string. In the provided examples, it is assumed that the pathPrefix is __health.)

  • __gtg - the GoodToGoo endpoint
    • params:
      • categories - the healthcheck will be performed on the services belonging to the provided categories.
      • cache - if set to false, the healthchecks will be performed without the help of cache. By default, the cache is used.
    • returns a 503 Service Unavailable status code in the following cases:
      • if at least one of the provided categories is disabled (see sticky functionality)
      • if at least one of the checked services is unhealthy
    • returns a 200 OK status code otherwise
    • example: localhost:8080/__gtg?cache=false&categories=read,publish
  • <pathPrefix>/__health or simply <pathPrefix> - Perform services healthcheck.
    • params:
      • categories - the healthcheck will be performed on the services belonging to the provided categories.
      • cache - if set to false, the healthchecks will be performed without the help of cache. By default, the cache is used.
    • example: localhost:8080/__health?cache=false&categories=read,publish
  • <pathPrefix>/__pods-health - Perform pods healthcheck for a service.
    • params:
      • service-name - The healthcheck will be performed only for pods belonging to the provided service.
    • example: localhost:8080/__health/__pods-health?service-name=api-policy-component
  • <pathPrefix>/__pod-individual-health - Retrieves the healthchecks of the app running inside the pod.
    • params:
      • pod-name - The name of the pod for which the healthchecks will be retrieved.
    • example: localhost:8080/__health/__pod-individual-health?pod-name=api-policy-component2912-12341
  • <pathPrefix>/add-ack - (POST) Acknowledges a service
    • params:
      • service-name - The service to be acknowledged.
    • example: localhost:8080/__health/add-ack?service-name=api-policy-component (request body: ack-msg=this is the message for ack)
    • request body: ack-msg the acknowledge message.
  • <pathPrefix>/rem-ack - Removes the acknowledge of a service
    • params:
      • service-name - The service to be updated.
    • example: localhost:8080/__health/rem-ack?service-name=api-policy-component
  • <pathPrefix>/enable-category - Enables a category. This is used for sticky categories which are unhealthy.
    • params:
      • category-name - The category to be enabled.
    • example: localhost:8080/__health/enable-category?category-name=read
  • <pathPrefix>/disable-category - Disables a category. This is useful when doing a failover.
    • params:
      • category-name - The category to be disabled.
    • example: localhost:8080/__health/disable-category?category-name=read

Admin endpoints

  • __health

  • __gtg

Call sequence

main.go call sequence

  • controller.go: initializeController
    • calls service.go: initializeHealthCheckService
  • prometheusFeeder.go: newPrometheusFeeder & feed
  • listen starts HTTP server
    • path / -> handler.go: handleServicesHealthCheck
    • path /__pods-health -> handler.go: handlePodsHealthCheck
    • path /__pod-individual-health -> handler.go: handleIndividualPodHealthCheck

httpHandler methods

  • handler.go
    • handleServicesHealthCheck
      • controller.go: buildServicesHealthResult
      • sort, format and return healthcheck results
      • buildServicesCheckHTMLResponse (format result of buildServicesHealthResult for output)

healthCheckController methods

  • controller.go
    • buildServicesHealthResult
      • getMatchingCategories (filter categories)
      • if useCache -> cachingController.go: collectChecksFromCachesFor
      • if NOT useCache -> runServiceChecksFor
      • runServiceChecksByServiceNames
        • runServiceChecksByServiceNames
          • get services list from k8sHealthcheckService.getServicesMapByNames
          • runServiceChecksByServiceNames
            • k8sHealthcheckService.getDeployments
            • checks health for all services using go-fthealth.RunCheck
            • loops through the result in parallel and calculates the severity of each service using severityController.go: getSeverityForService
            • loops through the services list and updates acks using updateHealthCheckWithAckMsg
            • cachingController.go: updateCachedHealth
              • schedules recurring checks for services not already in measuredServices
              • in practice this means scheduling checks only on startup
      • disableStickyFailingCategories
  • cachingController.go
  • severityController.go
    • getSeverityForService
      • k8sHealthcheckService.getPodsForService

k8sHealthcheckService methods

  • service.go
    • initializeHealthCheckService
      • starts as a Go routine watchAcks (load and update the service acks)
      • starts as a Go routine watchServices (load and update the service list)
    • watchServices
      • using the k8s API gets all services matching kubectl get services -l hasHealthcheck=true
      • prepares them as service structures and saves into the k8sHealthcheckService.services.m map
      • after all services are processed it logs Services watching terminated. Reconnecting... and invokes itself again
    • watchAcks
      • using the k8s API gets all (should be only one currently) configmaps matching kubectl get configmaps -l healthcheck-acknowledgements-for=aggregate-healthcheck
      • updates the service.ack key of the k8sHealthcheckService.services.m map
      • after all acks are processed it logs Acks configMap watching terminated. Reconnecting..." and invokes itself again
    • getCategories
      • using the k8s API gets all configmaps matching kubectl get configmaps -l healthcheck-categories-for=aggregate-healthcheck
    • getDeployments
      • using the k8s API gets all deployment and statefulset names along with their desired replica count
    • getPodsForService
      • using the k8s API gets all pods matching kubectl get pods -l app=%s