tsuru.yaml is a special file located in the root of the application. The name of
the file may be
This file is used to describe certain aspects of your app. Currently it describes information about deployment hooks and deployment time health checks. How to use this features is described below.
tsuru provides some deployment hooks, like
build. Deployment hooks allow developers to run commands before and after
Here is an example about how to declare this hooks in your tsuru.yaml file:
hooks: restart: before: - python manage.py generate_local_file after: - python manage.py clear_local_cache build: - python manage.py collectstatic --noinput - python manage.py compress
tsuru supports the following hooks:
restart:before: this hook lists commands that will run before the unit is restarted. Commands listed in this hook will run once per unit. For instance, imagine there’s an app with two units and the
tsuru.yamlfile listed above. The command python manage.py generate_local_file would run two times, once per unit.
restart:after: this hook is like before-each, but runs after restarting a unit.
build: this hook lists commands that will be run during deploy, when the image is being generated.
You can declare a health check in your tsuru.yaml file. This health check will be called during the deployment process and tsuru will make sure this health check is passing before continuing with the deployment process.
If tsuru fails to run the health check successfully it will abort the deployment
before switching the router to point to the new units, so your application will
never be unresponsive. You can configure the maximum time to wait for the
application to respond with the
Here is how you can configure a health check in your yaml file:
healthcheck: path: /healthcheck scheme: http method: GET status: 200 match: .*OKAY.* allowed_failures: 0 use_in_router: false router_body: content
healthcheck:path: Which path to call in your application. This path will be called for each unit. It is the only mandatory field, if it’s not set your health check will be ignored.
healthcheck:scheme: Which scheme to use. Defaults to http.
healthcheck:method: The method used to make the http request. Defaults to GET.
healthcheck:status: The expected response code for the request. Defaults to 200. This field is ignored in
kubernetesprovisioner, which always expects a status code greater than or equal to 200 and less than 400.
healthcheck:match: A regular expression to be matched against the request body. If it’s not set the body won’t be read and only the status code will be checked. This regular expression uses Go syntax and runs with
healthcheck:allowed_failures: The number of allowed failures before that the health check consider the application as unhealthy. Defaults to 0.
healthcheck:use_in_router: Whether this health check path should also be registered in the router. Please, ensure that the check is consistent to prevent units being disabled by the router. Defaults to false. When an app has no explicit healthcheck or use_in_router is false a default healthcheck is configured.
healthcheck:router_body: body passed to the router when
healthcheck:timeout_seconds: The timeout for each healthcheck call in seconds. Defaults to 60 seconds.
healthcheck:interval_seconds: Exclusive to the
kubernetesprovisioner. The interval in seconds between each active healthcheck call if
use_in_routeris set to true. Defaults to 10 seconds.
healthcheck:force_restart: Exclusive to the
kubernetesprovisioner. Whether the unit should be restarted after
allowed_failuresconsecutive healthcheck failures. (Sets the liveness probe in the Pod.)
Kubernetes specific configs¶
If your app is running on a Kubernetes provisioned pool, you can set specific configurations for Kubernetes. These configurations will be ignored if your app is running on another provisioner.
You can configure which ports will be exposed on each process of your app. Here’s a complete example:
kubernetes: groups: pod1: process1: ports: - name: main-port protocol: tcp target_port: 4123 port: 8080 - name: other-port protocol: udp port: 5000 pod2: process2:
groups key you can list each pod name - currently tsuru only supports
one process per pod -, and inside each one, the processes names.
For each process, you can configure each exposed port, in
kubernetes:groups:<group>:<process>:ports:name: A descriptive name for the port. This field is optional.
kubernetes:groups:<group>:<process>:ports:protocol: The port protocol. The accepted values are
kubernetes:groups:<group>:<process>:ports:target_port: The port that the process is listening on. If omitted,
portvalue will be used.
kubernetes:groups:<group>:<process>:ports:port: The port that will be exposed on a Kubernetes service. If omitted,
target_portvalue will be used.
target_port are omitted in a port config, the deploy
You can set a process to expose no ports (like a worker, for example) with an
empty field, like
The configuration for multiple ports still has a couple of limitations:
- healthcheck will be set to use the first configured port in each process
- only the first port of the web process (or the only process, in case there’s
only one) will be exposed in the router - but you can access the other ports
from other apps in the same cluster, using
Kubernetes DNS records,