tsuru.yaml¶

tsuru.yaml is a special file located in the root of the application. The name of the file may be tsuru.yaml or tsuru.yml.

This file is used to describe certain aspects of your app. Currently, it describes information about deployment hooks and deployment time health checks. The use of these features is described bellow.

Deployment hooks¶

tsuru provides some deployment hooks, like restart:before, restart:after and build. Deployment hooks allow developers to run commands before and after some commands.

An example on how to declare these hooks in tsuru.yaml file is described bellow:

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.yaml file 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 deployment when the image is being generated.

Healthcheck¶

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 docker:healthcheck:max-time config.

Health checks may also be used by the application router and by kubernetes, so you must ensure that the check is consistent to prevent units from being temporarily removed from the router.

Example on how you can configure a HTTP based health check in your yaml file:

healthcheck:
  path: /healthcheck
  scheme: http
  headers:
    Host: test.com
    X-Custom-Header: xxx
  allowed_failures: 0
  interval_seconds: 10
  timeout_seconds: 60
  deploy_timeout_seconds: 180

  // Ignored in Kubernetes provisioner pools:
  method: GET
  status: 200
  use_in_router: false
  match: .*OKAY.*
  router_body: content

Example of a command based healthcheck (Kubernetes only):

healthcheck:
  command: ["curl", "-f", "-XPOST", "http://localhost:8888"]

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. This field is ignored in kubernetes provisioner, GET is always used. Defaults to GET.
healthcheck:status: The expected response code for the request. Defaults to 200. This field is ignored in kubernetes provisioner, which always expects a status code greater than or equal to 200 and less than 400.
healthcheck:headers: Additional headers to use for the request. Headers name should be capitalized. It is optional.
healthcheck:match: A regular expression to be matched against the request body. This field is ignored in kubernetes provisioner, use healthcheck:command if a more complex healthcheck is necessary. 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 . matching \n (s flag).
healthcheck:allowed_failures: The number of allowed failures before that the health check consider the application as unhealthy. Defaults to 3 on kubernetes pools and 0 on docker pools.
healthcheck:use_in_router: Whether this health check path should also be registered in the router. This field is ignored in kubernetes provisioner, which constantly calls the healthcheck every interval_seconds. Defaults to false in other provisioners. When an app has no explicit healthcheck or use_in_router is false a default healthcheck is configured.
healthcheck:router_body: Body content passed to the router when use_in_router is true. This field is ignored in kubernetes provisioner, use healthcheck:command if a more complex healthcheck is necessary.
healthcheck:timeout_seconds: The timeout for each healthcheck call in seconds. Defaults to 60 seconds.
healthcheck:deploy_timeout_seconds: The timeout for the first successful healthcheck response after the application process has started during a new deploy. During this time a new healthcheck attempt will be made every healthcheck:interval_seconds. If the healthcheck is not successful in this time the deploy will be aborted and rolled back. Defaults to max-time global config.
healthcheck:command: Exclusive to the kubernetes provisioner. A command to execute inside the unit container. Exit status of zero is considered healthy and non-zero is unhealthy. This option defaults to an empty string array. If healthcheck:path is set, this option will be ignored.
healthcheck:interval_seconds: Exclusive to the kubernetes provisioner. The interval in seconds between each active healthcheck call if use_in_router is set to true. Defaults to 10 seconds.
healthcheck:force_restart: Exclusive to the kubernetes provisioner. Whether the unit should be restarted after allowed_failures consecutive 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:

Inside 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 ports key:

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 TCP (default) and UDP.
kubernetes:groups:<group>:<process>:ports:target_port: The port that the process is listening on. If omitted, port value will be used.
kubernetes:groups:<group>:<process>:ports:port: The port that will be exposed on a Kubernetes service. If omitted, target_port value will be used.

If both port and target_port are omitted in a port config, the deploy will fail.

You can set a process to expose no ports (like a worker, for example) with an empty field, like process2 above.

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, like appname-processname.namespace.svc.cluster.local