watchdog

README

Watchdog

The watchdog provides an unmanaged and generic interface between the outside world and your function. Its job is to marshal a HTTP request accepted on the API Gateway and to invoke your chosen application. The watchdog is a tiny Golang webserver - see the diagram below for how this process works.

Above: a tiny web-server or shim that forks your desired process for every incoming HTTP request

Every function needs to embed this binary and use it as its ENTRYPOINT or CMD, in effect it is the init process for your container. Once your process is forked the watchdog passses in the HTTP request via stdin and reads a HTTP response via stdout. This means your process does not need to know anything about the web or HTTP.

Create a new function the easy way

Create a function via the CLI

The easiest way to create a function is to use a template and the FaaS CLI. The CLI allows you to abstract all Docker knowledge away, you just have to write a handler file in one of the supported programming languages.

• Your first serverless Python function with OpenFaaS

• Read a tutorial on the FaaS CLI

Delve deeper

Package your function

Here's how to package your function if you don't want to use the CLI or have existing binaries or images:

• Use an existing or a new Docker image as base image FROM
• Add the fwatchdog binary from the Releases page via curl or ADD https://
• Set an fprocess (function process) environmental variable with the function you want to run for each request
• Expose port 8080
• Set the CMD to fwatchdog

Example Dockerfile for an echo function:

FROM alpine:3.7

ADD https://github.com/openfaas/faas/releases/download/0.8.0/fwatchdog /usr/bin
RUN chmod +x /usr/bin/fwatchdog

# Define your binary here
ENV fprocess="/bin/cat"

CMD ["fwatchdog"]

Implementing a health-check

At any point in time, if you detect that your function has become unhealthy and needs to restart, then you can delete the /tmp/.lock file which invalidates the check and causes Swarm to re-schedule the function.

• Kubernetes

For Kubernetes the health check is added through automation without you needing to alter the Dockerfile.

• Swarm

A Docker Swarm Healthcheck is required and is best practice. It will make sure that the watchdog is ready to accept a request before forwarding requests via the API Gateway. If the function or watchdog runs into an unrecoverable issue Swarm will also be able to restart the container.

Here is an example of the echo function implementing a healthcheck with a 5-second checking interval.

FROM functions/alpine

ENV fprocess="cat /etc/hostname"

HEALTHCHECK --interval=5s CMD [ -e /tmp/.lock ] || exit 1

The watchdog process creates a .lock file in /tmp/ on starting its internal Golang HTTP server. [ -e file_name ] is shell to check if a file exists. With Windows Containers this is an invalid path so you may want to set the suppress_lock environmental variable.

Read my Docker Swarm tutorial on Healthchecks:

• Test-drive Docker Healthcheck in 10 minutes

Environmental overrides:

The watchdog can be configured through environmental variables. You must always specifiy an fprocess variable.

Option	Usage
fprocess	The process to invoke for each function call (function process). This must be a UNIX binary and accept input via STDIN and output via STDOUT
cgi_headers	HTTP headers from request are made available through environmental variables - Http_X_Served_By etc. See section: Handling headers for more detail. Enabled by default
marshal_request	Instead of re-directing the raw HTTP body into your fprocess, it will first be marshalled into JSON. Use this if you need to work with HTTP headers and do not want to use environmental variables via the cgi_headers flag.
content_type	Force a specific Content-Type response for all responses
write_timeout	HTTP timeout for writing a response body from your function (in seconds)
read_timeout	HTTP timeout for reading the payload from the client caller (in seconds)
suppress_lock	The watchdog will attempt to write a lockfile to /tmp/ for swarm healthchecks - set this to true to disable behaviour.
exec_timeout	Hard timeout for process exec'd for each incoming request (in seconds). Disabled if set to 0
write_debug	Write all output, error messages, and additional information to the logs. Default is false
combine_output	True by default - combines stdout/stderr in function response, when set to false stderr is written to the container logs and stdout is used for function response

Advanced / tuning

(New) of-watchdog and HTTP mode

• of-watchdog

Forking a new process per request has advantages such as process isolation, portability and simplicity. Any process can be made into a function without any additional code. The of-watchdog and its "HTTP" mode is an optimization which maintains one single process between all requests.

A new version of the watchdog is being tested over at openfaas-incubator/of-watchdog.

This re-write is mainly structural for on-going maintenance. It will be a drop-in replacement for the existing watchdog and also has binary releases available.

Working with HTTP headers

Headers and other request information are injected into environmental variables in the following format:

The X-Forwarded-By header becomes available as Http_X_Forwarded_By

• Http_Method - GET/POST etc
• Http_Query - QueryString value
• Http_ContentLength - gives the total content-length of the incoming HTTP request received by the watchdog.

This behaviour is enabled by the cgi_headers environmental variable which is enabled by default.

Here's an example of a POST request with an additional header and a query-string.

$ cgi_headers=true fprocess=env ./watchdog &
2017/06/23 17:02:58 Writing lock-file to: /tmp/.lock

$ curl "localhost:8080?q=serverless&page=1" -X POST -H X-Forwarded-By:http://my.vpn.com

This is what you'd see if you had set your fprocess to env on a Linux system:

Http_User_Agent=curl/7.43.0
Http_Accept=*/*
Http_X_Forwarded_By=http://my.vpn.com
Http_Method=POST
Http_Query=q=serverless&page=1

You can also use the GET verb:

$ curl "localhost:8080?action=quote&qty=1&productId=105"

The output from the watchdog would be:

Http_User_Agent=curl/7.43.0
Http_Accept=*/*
Http_Method=GET
Http_Query=action=quote&qty=1&productId=105

You can now use HTTP state from within your application to make decisions.

HTTP methods

The HTTP methods supported for the watchdog are:

With a body:

• POST, PUT, DELETE, UPDATE

Without a body:

• GET

The API Gateway currently supports the POST route for functions.

Content-Type of request/response

By default the watchdog will match the response of your function to the "Content-Type" of the client.

• If your client sends a JSON post with a Content-Type of application/json this will be matched automatically in the response.
• If your client sends a JSON post with a Content-Type of text/plain this will be matched automatically in the response too

To override the Content-Type of all your responses set the content_type environmental variable.

I don't want to use the watchdog

This is an unsupported use-case for the OpenFaaS project however if your container conforms to the requirements below then the OpenFaaS API gateway and other tooling will manage and scale your service.

You will need to provide a lock-file at /tmp/.lock so that the orchestration system can run healthchecks on your container. If you are using Docker Swarm make sure you provide a HEALTHCHECK instruction in your Dockerfile - samples are given in the faas repository.

• Expose TCP port 8080 over HTTP
• Create /tmp/.lock or in whatever location responds to the OS tempdir syscall

Tuning auto-scaling

Auto-scaling starts at 1 replica and steps up in blocks of 5:

• 1->5
• 5->10
• 10->15
• 15->20

You can override the minimum and maximum scale of a function through labels.

Add these labels to the deployment if you want to sacle between 2 and 15 replicas.

com.openfaas.scale.min: "2"
com.openfaas.scale.max: "15"

The labels are optional.

Disabling auto-scaling

If you want to disable auto-scaling for a function then set the minimum and maximum scale to the same value i.e. "1".

As an alternative you can also remove AlertManager or scale it to 0 replicas.