Kubernetes::Health
This gem opens a dedicated HTTP port to allow Kubernetes to monitor your Rails app while it is running migrations, Rake tasks, Sidekiq, or Puma.
Features
- Puma and Sidekiq metrics for autoscaling.
- Prometheus and JSON metrics.
- add routes
/_readiness,/_livenesson Rails Stack. - add routes
/_readiness,/_livenessand/_metricsas a puma plugin at another port to avoid problems when your app get busy. (code copied frompuma-metricsgem). - add routes
/_readinessand/_livenesswhilerake db:migrateruns. (optional) - add routes
/_metricswhilesidekiqruns. (optional) - add support to avoid parallel running of
rake db:migratewhile keep kubernetes waiting (PostgreSQL required). - allow custom checks for
/_readinessand/_liveness.
Installation
Add this line to your application's Gemfile:
gem 'kubernetes-health', '~> 3.12.1'
Enabling puma plugin
add in config/puma.rb
plugin 'kubernetes'
kubernetes_url 'tcp://0.0.0.0:9393'
In Kubernetes you need to configure your deployment readinessProbe and livenessProbe like this:
livenessProbe:
httpGet:
path: /_liveness
port: 9393
initialDelaySeconds: 30
timeoutSeconds: 5
failureThreshold: 3
successThreshold: 1
readinessProbe:
httpGet:
path: /_readiness
port: 9393
initialDelaySeconds: 30
timeoutSeconds: 5
failureThreshold: 3
successThreshold: 1
Setting failureThreshold is import to avoid problems when app finish migrates and is starting the web process.
Enabling liveness/readiness routes while rake db:migrate runs
Your Dockerfile's entry script needs to run migrates before start your web app.
Add KUBERNETES_HEALTH_ENABLE_RACK_ON_MIGRATE=true environment variable.
or add in your application.rb.
# default: false
Kubernetes::Health::Config.enable_rack_on_migrate = true
The defined port at config/puma.rb will be used but can be overrided by KUBERNETES_HEALTH_METRICS_PORT env var.
Enabling liveness/readiness routes while any rake tasks runs
If you need to run another rake tasks than db:migrate, like assets:precompile, you can enable the monitoring routes by this way:
Add a rake file enhancing the original task by using kubernetes_health:rack_on_rake task. For example:
# File: lib/tasks/kubernetes_health_enable_rack_on_assets_precompile.rake
Rake::Task['assets:precompile'].enhance(['kubernetes_health:rack_on_rake'])
# File: lib/tasks/kubernetes_health_enable_rack_on_assets_clobber.rake
Rake::Task['assets:clobber'].enhance(['kubernetes_health:rack_on_rake'])
I do recommend doing some checks to make it only enabled in the K8S environment.
The defined port at config/puma.rb will be used but can be overrided by KUBERNETES_HEALTH_METRICS_PORT env var.
Enabling liveness/readiness routes for sidekiq
Add KUBERNETES_HEALTH_ENABLE_RACK_ON_SIDEKIQ=true environment variable.
or add in your application.rb.
# default: false
Kubernetes::Health::Config.enable_rack_on_sidekiq = true
The defined port at config/puma.rb will be used but can be overrided by KUBERNETES_HEALTH_METRICS_PORT env var.
How rake and sidekiq monitoring works
It will run a rack server for /_readiness, /_liveness and /_metrics for rake and /_metrics for Sidekiq.
The liveness route will respond using 200 but readiness 503.
Avoiding migrations running in parallel and making kubernetes happy.
Rails already avoid migrations running in parallel, but it raises exceptions. This gem will just wait for other migrations without exit.
If you enable rack_on_migrate together with this, kubernetes will just wait, avoiding erros.
Add KUBERNETES_HEALTH_ENABLE_LOCK_ON_MIGRATE=true environment variable.
or add in your application.rb.
# default: false
Kubernetes::Health::Config.enable_lock_on_migrate = true
Customizing locking
By default it is working for PostgreSQL, but you can customize it using a lambda:
Kubernetes::Health::Config.lock_or_wait = lambda {
ActiveRecord::Base.connection.execute "SET lock_timeout TO '3600s'; SELECT pg_advisory_lock(123456789123456789);"
}
Kubernetes::Health::Config.unlock = lambda {
ActiveRecord::Base.connection.execute 'select pg_advisory_unlock(123456789123456789);'
}
Customizing checks
It only works for routes in rails stack, they are not executed while rake db:migrate runs.
I prefer do nothing else on liveness to avoid unnecessary CrashLoopBackOff status. params is optional (request params).
Kubernetes::Health::Config.live_if = lambda { |params|
true
}
Ex. Check if PostgreSQL is reachable on readiness indicating that credentials are setup right and keeps cache to avoid doing it a lot. params is optional (request params).
Kubernetes::Health::Config.ready_if = lambda { |params|
return $kubernetes_health_test_db_connection if $kubernetes_health_test_db_connection
$kubernetes_health_test_db_connection = ActiveRecord::Base.connection.execute("SELECT 1").cmd_tuples == 1
}
Customizing routes
Kubernetes::Health::Config.route_liveness = '/liveness'
Kubernetes::Health::Config.route_readiness = '/readiness'
Kubernetes::Health::Config.route_metrics = '/metrics'
or using env
KUBERNETES_HEALTH_LIVENESS_ROUTE='/liveness'
KUBERNETES_HEALTH_READINESS_ROUTE='/readiness'
KUBERNETES_HEALTH_RESPONSE_FORMAT='/metrics'
Response format
If you are using https://github.com/zalando-incubator/kube-metrics-adapter you will want to use json format.
Default is prometheus.
Kubernetes::Health::Config.response_format = 'json'
or using env
KUBERNETES_HEALTH_RESPONSE_FORMAT=json
Customizing requests logs
Kubernetes::Health::Config.request_log_callback = lambda { |req, http_code, content|
Rails.logger.debug "Kubernetes Health: Rack on Migrate - Request: Path: #{req.path_info} / Params: #{req.params} / HTTP Code: #{http_code}\n#{content}" rescue nil
}