Managing Application Processes¶

Deis Workflow manages your application as a set of processes that can be named, scaled and configured according to their role. This gives you the flexibility to easily manage the different facets of your application. For example, you may have web-facing processes that handle HTTP traffic, background worker processes that do async work, and a helper process that streams from the Twitter API.

By using a Procfile, either checked in to your application or provided via the CLI you can specify the name of the type and the application command that should run. To spawn other process types, use deis scale <type>=<n> to scale those types accordingly.

Default Process Types¶

In the absence of a Procfile, a single, default process type is assumed for each application.

Applications built using Buildpacks via git push implicitly receive a web process type, which starts the application server. Rails 4, for example, has the following process type:

web: bundle exec rails server -p $PORT

All applications utilizing Dockerfiles have an implied cmd process type, which runs the Dockerfile's CMD directive unmodified:

$ cat Dockerfile
FROM centos:latest
COPY . /app
WORKDIR /app
CMD python -m SimpleHTTPServer 5000
EXPOSE 5000

For the above Dockerfile-based application, the cmd process type would run the Docker CMD of python -m SimpleHTTPServer 5000.

Applications utilizing remote Docker images, a cmd process type is also implied, and runs the CMD specified in the Docker image.

Declaring Process Types¶

If you use Buildpack or Dockerfile builds and want to override or specify additional process types, simply include a file named Procfile in the root of your application's source tree.

The format of a Procfile is one process type per line, with each line containing the command to invoke:

<process type>: <command>

The syntax is defined as:

• <process type> – a lowercase alphanumeric string, is a name for your command, such as web, worker, urgentworker, clock, etc.
• <command> – a command line to launch the process, such as rake jobs:work.

This example Procfile specifies two types, web and sleeper. The web process launches a web server on port 5000 and a simple process which sleeps for 900 seconds and exits.

$ cat Procfile
web: bundle exec ruby web.rb -p ${PORT:-5000}
sleeper: sleep 900

If you are using remote Docker images, you may define process types by either running deis pull with a Procfile in your working directory, or by passing a stringified Procfile to the --procfile CLI option.

For example, passing process types inline:

$ deis pull deis/example-go:latest --procfile="cmd: /app/bin/boot"

Read a Procfile in another directory:

$ deis pull deis/example-go:latest --procfile="$(cat deploy/Procfile)"

Or via a Procfile located in your current, working directory:

$ cat Procfile
cmd: /bin/boot
sleeper: echo "sleeping"; sleep 900


$ deis pull -a steely-mainsail deis/example-go
Creating build... done

$ deis scale sleeper=1 -a steely-mainsail
Scaling processes... but first, coffee!
done in 0s
=== steely-mainsail Processes
--- cmd:
steely-mainsail-cmd-3291896318-nyrim up (v3)
--- sleeper:
steely-mainsail-sleeper-3291896318-oq1jr up (v3)

To remove a process type simply scale it to 0:

$ deis scale sleeper=0 -a steely-mainsail
Scaling processes... but first, coffee!
done in 3s
=== steely-mainsail Processes
--- cmd:
steely-mainsail-cmd-3291896318-nyrim up (v3)

Scaling Processes¶

Applications deployed on Deis Workflow scale out via the process model. Use deis scale to control the number of containers that power your app.

$ deis scale cmd=5 -a iciest-waggoner
Scaling processes... but first, coffee!
done in 3s
=== iciest-waggoner Processes
--- cmd:
iciest-waggoner-web-3291896318-09j0o up (v2)
iciest-waggoner-web-3291896318-3r7kp up (v2)
iciest-waggoner-web-3291896318-gc4xv up (v2)
iciest-waggoner-web-3291896318-lviwo up (v2)
iciest-waggoner-web-3291896318-kt7vu up (v2)

If you have multiple process types for your application you may scale the process count for each type separately. For example, this allows you to manage web process independently from background workers. For more information on process types see our documentation for Managing App Processes.

In this example, we are scaling the process type web to 5 but leaving the process type background with one worker.

$ deis scale web=5
Scaling processes... but first, coffee!
done in 4s
=== scenic-icehouse Processes
--- web:
scenic-icehouse-web-3291896318-7lord up (v2)
scenic-icehouse-web-3291896318-jn957 up (v2)
scenic-icehouse-web-3291896318-rsekj up (v2)
scenic-icehouse-web-3291896318-vwhnh up (v2)
scenic-icehouse-web-3291896318-vokg7 up (v2)
--- background:
scenic-icehouse-web-3291896318-background-yf8kh up (v2)

Scaling a process down, by reducing the process count, sends a TERM signal to the processes, followed by a SIGKILL if they have not exited within 30 seconds. Depending on your application, scaling down may interrupt long-running HTTP client connections.

For example, scaling from 5 processes to 3:

$ deis scale web=3
Scaling processes... but first, coffee!
done in 1s
=== scenic-icehouse Processes
--- background:
scenic-icehouse-web-3291896318-background-yf8kh up (v2)
--- web:
scenic-icehouse-web-3291896318-7lord up (v2)
scenic-icehouse-web-3291896318-rsekj up (v2)
scenic-icehouse-web-3291896318-vokg7 up (v2)

Autoscale¶

Autoscale allows adding a minimum and maximum number of pods on a per process type basis. This is accomplished by specifying a target CPU usage across all available pods.

This feature is built on top of Horizontal Pod Autoscaling in Kubernetes or HPA for short.

$ deis autoscale:set web --min=3 --max=8 --cpu-percent=75
Applying autoscale settings for process type web on scenic-icehouse... done

And then review the scaling rule that was created for web

$ deis autoscale:list
=== scenic-icehouse Autoscale

--- web:
Min Replicas: 3
Max Replicas: 8
CPU: 75%

Remove scaling rule

$ deis autoscale:unset web
Removing autoscale for process type web on scenic-icehouse... done

For autoscaling to work CPU requests have to be specified on each application Pod (can be done via deis limits --cpu). This allows the autoscale policies to do the appropriate calculations and make decisions on when to scale up and down.

Scale up can only happen if there was no rescaling within the last 3 minutes. Scale down will wait for 5 minutes from the last rescaling. That information and more can be found at HPA algorithm page.

Web vs Cmd Process Types¶

When deploying to Deis Workflow using a Heroku Buildpack, Workflow boots the web process type to boot the application server. When you deploy an application that has a Dockerfile or uses Docker images, Workflow boots the cmd process type. Both act similarly in that they are exposed to the router as web applications. However, the cmd process type is special because, if left undefined, it is equivalent to running the container without any additional arguments. (i.e. The process specified by the Dockerfile or Docker image's CMD directive will be used.)

If migrating an application from Heroku Buildpacks to a Docker-based deployment, Workflow will not automatically convert the web process type to cmd. To do this, you'll have to manually scale down the old process type and scale the new process type up.

Restarting an Application Processes¶

If you need to restart an application process, you may use deis ps:restart. Behind the scenes, Deis Workflow instructs Kubernetes to terminate the old process and launch a new one in its place.

$ deis ps
=== scenic-icehouse Processes
--- web:
scenic-icehouse-web-3291896318-7lord up (v2)
scenic-icehouse-web-3291896318-rsekj up (v2)
scenic-icehouse-web-3291896318-vokg7 up (v2)
--- background:
scenic-icehouse-background-3291896318-yf8kh up (v2)
$ deis ps:restart scenic-icehouse-background-3291896318-yf8kh
Restarting processes... but first, coffee!
done in 6s
=== scenic-icehouse Processes
--- background:
scenic-icehouse-background-3291896318-yd87g up (v2)

Notice that the process name has changed from scenic-icehouse-background-3291896318-yf8kh to scenic-icehouse-background-3291896318-yd87g. In a multi-node Kubernetes cluster, this may also have the effect of scheduling the Pod to a new node.