Customizing service life-cycle management

The health option allows you to specify different types of checks on your containers - readiness checks, liveness checks, and startup checks. These checks define a set of rules that determine whether your application is currently healthy. For instance, you can check whether an application is responding on an HTTP endpoint, or if a post-initialization file is present.

Readiness checks test whether newly started containers are ready to replace old containers. Until the new containers are ready, the old containers will not be killed, and the new containers will not be served traffic. This effectively provides zero down-time deployments.

Liveness checks continuously monitor your application while it is running. If your application fails a liveness check, it will be restarted. This is useful for issues that cannot be resolved otherwise.

Startup checks detect when a container has started. An active startup check will disable liveness and readiness checks until it succeeds. This prevents other checks from interrupting application startup. This is particularly useful for slow starting containers, because otherwise liveness checks might cause them to be killed before they are up and running. Requires Kubernetes v1.16 or greater.

Please see the official Kubernetes documentation regarding these checks.

The options below are available for all health checks. Note that you aren't required to configure all the options. Any options not configured will use their default values.

• type (string, defaults to 'http'): Accepted values are http, https, tcp, and exec.
• endpoint (string, defaults to '/'): The endpoint tested for status. Only for the http, and the https types.
• command (string, no default): The command executed to test for status. Must return exit-code 0. Only for the exec type.
• timeout (integer, defaults to 5): Maximum time in seconds to wait for a health check to complete.
• success_threshold (integer, defaults to 1): Number of consecutive successes to be considered healthy.
• failure_threshold (integer, defaults to 1): Number of consecutive failures to be considered unhealthy.
• initial_delay (integer, defaults to 1): Time in seconds to wait after container has started before starting liveness checks.
• period (integer, defaults to 5): Number of seconds between each consecutive health check.
• port (integer, defaults to container port): The port to run the health check against. Only for the http, https, and tcp types.
• http_headers (array, defaults to []): Custom headers to add for HTTP traffic. Only for the http, and https types. Contains an array of hashes with the name and value keys, both of string type.

services:
    [service_name]:
        health:
            ready:
                type: exec
                command: 'cat /tmp/initialization_complete'
            alive:
                type: http
                endpoint: /healthz
                success_threshold: 2
                failure_threshold: 2
                initial_delay: 10
                period: 30
                http_headers:
                - name: 'X-ID-Header'
                  value: 'john-smith'
            startup:
                type: exec
                command: 'cat /tmp/app_started'                  

This is a signal that is sent to the existing containers of the service before the new containers are started during deployment. An example could be USR1 - but it depends on what your container is running as to which signals make sense.

services:
    [service_name]:
        pre_start_signal: USR1

This is a stop sequence that is executed on your running containers before they are shut down. It is a sequence of wait times and signals to send to the process. If the sequence completes and the container is still running, a force kill will be sent. For example:

services:
    [service_name]:
        pre_stop_sequence: 1m:USR2:30s:USR1:50s

The example above, we'll wait 1 minute before sending the USR2 signal, then wait 30 seconds before sending the USR1 signal, and then wait 50 seconds before we force a kill. These are some examples of duration values that stop_grace and pre_stop_sequence can use - 1m (1 minute), 30s (30 seconds) and 1h (1 hour).

Valid time values are s for seconds, m for minutes and h for hours. Valid signal values for a signal are (without the quotes):

'ABRT', 'ALRM', 'BUS', 'CHLD', 'CONT', 'FPE', 'HUP', 'ILL', 'INT', 'IO', 'IOT', 'KILL', 'PIPE', 'PROF', 'QUIT', 'SEGV', 'STOP', 'SYS', 'TERM', 'TRAP', 'TSTP', 'TTIN', 'TTOU', 'URG', 'USR1', 'USR2', 'VTALRM', 'WINCH', 'XCPU', 'XFSZ'