mabuaisha / faas-chart

OpenFaaS - Serverless Functions Made Simple

OpenFaaS (Functions as a Service) is a framework for building serverless functions with Docker and Kubernetes which has first class support for metrics. Any process can be packaged as a function enabling you to consume a range of web events without repetitive boiler-plate coding.

Highlights

• Ease of use through UI portal and one-click install
• Write functions in any language for Linux or Windows and package in Docker/OCI image format
• Portable - runs on existing hardware or public/private cloud. Native Kubernetes support, Docker Swarm also available
• Operator / CRD option available
• faas-cli available with stack.yml for creating and managing functions
• Auto-scales according to metrics from Prometheus
• Scales to zero and back again and can be tuned at a per-function level
• Works with service-mesh
  • Tested with Istio including mTLS
  • Tested with Linkerd2 including mTLS

Deploy OpenFaaS

Note: If your cluster is not configured with role-based access control, then pass --set rbac=false. For further information, see Kubernetes: RBAC.

Note: If you can not use helm with Tiller, you can still use the helm CLI to generate custom YAML without server-side components.

---

Install

These instructions are for Intel (normal computers), jump to the end of the document for ARM and Raspberry Pi.

To use the chart, you will need Helm 2 or 3:

• Install helm

We recommend creating two namespaces, one for the OpenFaaS core services and one for the functions:

kubectl apply -f https://raw.githubusercontent.com/openfaas/faas-netes/master/namespaces.yml

You will now have openfaas and openfaas-fn. If you want to change the names or to install into multiple installations then edit namespaces.yml from the faas-netes repo.

Add the OpenFaaS helm chart:

helm repo add openfaas https://openfaas.github.io/faas-netes/

Now decide how you want to expose the services and edit the helm upgrade command as required.

• To use NodePorts (default) pass no additional flags
• To use a LoadBalancer add --set serviceType=LoadBalancer
• To use an IngressController add --set ingress.enabled=true

Note: even without a LoadBalancer or IngressController you can access your gateway at any time via kubectl port-forward.

Deploy

Note that the commands will differ slightly between versions, if not specified, the instructions are for helm 2.

Now deploy OpenFaaS from the helm chart repo:

helm repo update \
 && helm upgrade openfaas --install openfaas/openfaas \
    --namespace openfaas  \
    --set functionNamespace=openfaas-fn \
    --set generateBasicAuth=true 

The above command will also update your helm repo to pull in any new releases.

Retrieve the OpenFaaS credentials with:

PASSWORD=$(kubectl -n openfaas get secret basic-auth -o jsonpath="{.data.basic-auth-password}" | base64 --decode) && \
echo "OpenFaaS admin password: $PASSWORD"

Generate basic-auth credentials

The chart has a pre-install hook which can generate basic-auth credentials, enable it with --set generateBasicAuth=true.

Alternatively, you can set generateBasicAuth to false and generate or supply the basic-auth credentials yourself. This is the option you may want if you are using helm template.

# generate a random password	
PASSWORD=$(head -c 12 /dev/urandom | shasum| cut -d' ' -f1)	
kubectl -n openfaas create secret generic basic-auth \	
--from-literal=basic-auth-user=admin \	
--from-literal=basic-auth-password="$PASSWORD"	

Tuning cold-start

The concept of a cold-start in OpenFaaS only applies if you A) use faas-idler and B) set a specific function to scale to zero. Otherwise there is not a cold-start, because at least one replica of your function remains available.

There are two ways to reduce the Kubernetes cold-start for a pre-pulled image, which is around 1-2 seconds.

1. Don't set the function to scale down to zero, just set it a minimum availability i.e. 1/1 replicas
2. Use async invocations via the /async-function/<name> route on the gateway, so that the latency is hidden from the caller
3. Tune the readinessProbes to be aggressively low values. This will reduce the cold-start at the cost of more kubelet CPU usage

To achieve around 1s coldstart, set values.yaml:

faasnetes:

# redacted
  readinessProbe:
    initialDelaySeconds: 0
    timeoutSeconds: 1
    periodSeconds: 1
  livenessProbe:
    initialDelaySeconds: 0
    timeoutSeconds: 1
    periodSeconds: 1
# redacted
  imagePullPolicy: "IfNotPresent"    # Image pull policy for deployed functions

In addition:

• Pre-pull images on each node
• Use an in-cluster registry to reduce the pull latency for images
• Set the imagePullPolicy to IfNotPresent so that the kubelet only pulls images which are not already available
• Explore alternatives such as not scaling to absolute zero, and using async calls which do not show the cold start

httpProbe vs. execProbe

A note on health-checking probes for functions:

• httpProbe - (default) most efficient. (compatible with Istio >= 1.1.5)
• execProbe - least efficient option, but compatible with Istio < 1.1.5

Use --set faasnetes.httpProbe=true/false to toggle between http / exec probes.

Verify the installation

Once all the services are up and running, log into your gateway using the OpenFaaS CLI. This will cache your credentials into your ~/.openfaas/config.yml file.

Fetch your public IP or NodePort via kubectl get svc -n openfaas gateway-external -o wide and set it as an environmental variable as below:

export OPENFAAS_URL=http://127.0.0.1:31112

If using a remote cluster, you can port-forward the gateway to your local machine:

kubectl port-forward -n openfaas svc/gateway 31112:8080 &

Now log in with the CLI and check connectivity:

echo -n $PASSWORD | faas-cli login -g $OPENFAAS_URL -u admin --password-stdin

faas-cli version

OpenFaaS Operator and Function CRD

If you would like to work with Function CRDs there is an alternative controller to faas-netes named OpenFaaS Operator which can be swapped in at deployment time. The OpenFaaS Operator is suitable for development and testing and may replace the faas-netes controller in the future. The Operator is compatible with Kubernetes 1.9 or later.

To use it, add the flag: --set operator.create=true when installing with Helm.

faas-netes vs OpenFaaS Operator

The faas-netes controller is the most tested, stable and supported version of the OpenFaaS integration with Kubernetes. In contrast the OpenFaaS Operator is based upon the codebase and features from faas-netes, but offers a tighter integration with Kubernetes through CustomResourceDefinitions. This means you can type in kubectl get functions for instance.

See also: Introducing the OpenFaaS Operator

Deployment with helm template

This option is good for those that have issues with or concerns about installing Tiller, the server/cluster component of helm. Using the helm CLI, we can pre-render and then apply the templates using kubectl.

1. Clone the faas-netes repository

   git clone https://github.com/openfaas/faas-netes.git
   cd faas-netes

2. Render the chart to a Kubernetes manifest called openfaas.yaml

   Helm 3:

   helm template \
     openfaas chart/openfaas/ \
     --namespace openfaas \
     --set basic_auth=true \
     --set functionNamespace=openfaas-fn > openfaas.yaml

   Helm 2:

   helm template chart/openfaas \
       --name openfaas \
       --namespace openfaas  \
       --set basic_auth=true \
       --set functionNamespace=openfaas-fn > openfaas.yaml

   You can set the values and overrides just as you would in the install/upgrade commands above.

3. Install the components using kubectl

   kubectl apply -f namespaces.yml,openfaas.yaml

Now verify your installation.

Test a local helm chart

You can run the following command from within the faas-netes/chart folder in the faas-netes repo.

helm upgrade --install openfaas openfaas/openfaas \
    --namespace openfaas  \
    --set basic_auth=true \
    --set functionNamespace=openfaas-fn

Exposing services

NodePorts

By default a NodePort will be created for the API Gateway.

Metrics

You temporarily access the Prometheus metrics by ueing port-forward

kubectl --namespace openfaas port-forward deployment/prometheus 31119:9090

Then open http://localhost:31119 to directly query the OpenFaaS metrics scraped by Prometheus.

LB

If you're running on a cloud such as AKS or GKE you will need to pass an additional flag of --set serviceType=LoadBalancer to tell helm to create LoadBalancer objects instead. An alternative to using multiple LoadBalancers is to install an Ingress controller.

Deploy with an IngressController

In order to make use of automatic ingress settings you will need an IngressController in your cluster such as Traefik or Nginx.

Add --set ingress.enabled to enable ingress pass --set ingress.enabled=true when running the installation via helm.

By default services will be exposed with following hostnames (can be changed, see values.yaml for details):

• gateway.openfaas.local

Endpoint load-balancing

Some configurations in combination with client-side KeepAlive settings may because load to be spread unevenly between replicas of a function. If you experience this, there are three ways to work around it:

• Install Linkerd2 which takes over load-balancing from the Kubernetes L4 Service (recommended)

• Disable KeepAlive in the client-side code (not recommended)

• Configure the gateway to pass invocations through to the faas-netes provider (alternative to using Linkerd2)

  --set gateway.directFunctions=false

  In this mode, all invocations will pass through the gateway to faas-netes, which will look up endpoint IPs directly from Kubernetes, the additional hop may add some latency, but will do fair load-balancing, even with KeepAlive.

SSL / TLS

If you require TLS/SSL then please make use of an IngressController. A full guide is provided to enable TLS for the OpenFaaS Gateway using cert-manager and Let's Encrypt.

Istio mTLS

To install OpenFaaS with Istio mTLS pass --set istio.mtls=true and disable the HTTP probes:

helm upgrade openfaas --install chart/openfaas \
    --namespace openfaas  \
    --set basic_auth=true \
    --set functionNamespace=openfaas-fn \
    --set exposeServices=false \
    --set faasnetes.httpProbe=false \
    --set httpProbe=false \
    --set istio.mtls=true

The above command will enable mTLS for the openfaas control plane services and functions excluding NATS.

Note that the above instructions were tested on GKE 1.13 and Istio 1.2

Zero scale

Scale-up from zero (on by default)

Scaling up from zero replicas is enabled by default, to turn it off set scaleFromZero to false in the helm chart options for the gateway component.

--set gateway.scaleFromZero=true/false

Scale-down to zero (off by default)

Scaling down to zero replicas can be achieved either through the REST API and your own controller, or by using the faas-idler component.

By default the faas-idler is set to only do a dryRun and to not scale any functions down.

--set faasIdler.dryRun=true/false

The faas-idler will only scale down functions which have marked themselves as eligible for this behaviour through the use of a label: com.openfaas.scale.zero=true.

See also: faas-idler README.

Removing the OpenFaaS

All control plane components can be cleaned up with helm:

Helm 3:

helm delete openfaas --namespace openfaas

Helm 2:

helm delete --purge openfaas

Follow this by the following to remove all other associated objects:

kubectl delete namespace openfaas openfaas-fn

In some cases your additional functions may need to be either deleted before deleting the chart with faas-cli or manually deleted using kubectl delete.

ARM

If you would like to deploy OpenFaaS to ARM i.e. Raspberry Pi, ARM64 machines provided by Packet.net, Scaleway or to AWS Graviton, then you should use the appropriate values.yaml file.

• values-armhf.yml - for Raspberry Pi and other ARMv7 boards (run uname -a to find out which you have)
• values-arm64.yml - for everything else (arm64 or aarch64)

It is recommended that you install OpenFaaS to ARM machines using k3sup instead of helm directly since it will determine the correct values to be used.

See also: Kubernetes and Raspberry Pi in the docs

Getting help

Feel free to seek out help using the OpenFaaS Slack workspace, please do not raise issues for technical support, unless you suspect and can provide instructions for reproducing an error in the chart.

Configuration

Additional OpenFaaS options in values.yaml.

Parameter	Description	Default
functionNamespace	Functions namespace, preferred openfaas-fn	default
clusterRole	Set to true if you'd like to use multiple namespaces for functions	false
async	Enables asynchronous function invocations. If .nats.external.enabled is false, also deploys NATS Streaming	true
exposeServices	Expose NodePorts/LoadBalancer	true
serviceType	Type of external service to use NodePort/LoadBalancer	NodePort
basic_auth	Enable basic authentication on the Gateway	true
generateBasicAuth	Generate admin password for basic authentication	false
rbac	Enable RBAC	true
httpProbe	Setting to true will use HTTP for readiness and liveness probe on the OpenFaaS system Pods (compatible with Istio >= 1.1.5)	true
psp	Enable Pod Security Policy for OpenFaaS accounts	false
securityContext	Deploy with a securityContext set, this can be disabled for use with Istio sidecar injection	true
openfaasImagePullPolicy	Image pull policy for openfaas components, can change to IfNotPresent in offline env	Always
kubernetesDNSDomain	Domain name of the Kubernetes cluster	cluster.local
operator.create	Use the OpenFaaS operator CRD controller, default uses faas-netes as the Kubernetes controller	false
operator.createCRD	Create the CRD for OpenFaaS Function definition	true
ingress.enabled	Create ingress resources	false
faasnetes.httpProbe	Use a httpProbe instead of exec	false
ingressOperator.create	Create the ingress-operator component	false
ingressOperator.replicas	Replicas of the ingress-operator	1
ingressOperator.image	Container image used in ingress-operator	openfaas/ingress-operator:0.4.0
ingressOperator.resources	Limits and requests for memory and CPU usage	Memory Requests: 25Mi
faasnetes.readTimeout	Queue worker read timeout	60s
faasnetes.writeTimeout	Queue worker write timeout	60s
faasnetes.imagePullPolicy	Image pull policy for deployed functions	Always
faasnetes.setNonRootUser	Force all function containers to run with user id 12000	false
gateway.directFunctions	Invoke functions directly without using the provider	true
gateway.replicas	Replicas of the gateway, pick more than 1 for HA	1
gateway.readTimeout	Queue worker read timeout	65s
gateway.writeTimeout	Queue worker write timeout	65s
gateway.upstreamTimeout	Maximum duration of upstream function call, should be lower than readTimeout/writeTimeout	60s
gateway.scaleFromZero	Enables an intercepting proxy which will scale any function from 0 replicas to the desired amount	true
gateway.maxIdleConns	Set max idle connections from gateway to functions	1024
gateway.maxIdleConnsPerHost	Set max idle connections from gateway to functions per host	1024
gateway.logsProviderURL	Set a custom logs provider url	""
queueWorker.durableQueueSubscriptions	Whether to use a durable queue subscription	false
queueWorker.queueGroup	The name of the queue group used to process asynchronous function invocations	faas
queueWorker.replicas	Replicas of the queue-worker, pick more than 1 for HA	1
queueWorker.ackWait	Max duration of any async task/request	60s
nats.channel	The name of the NATS Streaming channel to use for asynchronous function invocations	faas-request
nats.external.clusterName	The name of the externally-managed NATS Streaming server	``
nats.external.enabled	Whether to use an externally-managed NATS Streaming server	false
nats.external.host	The host at which the externally-managed NATS Streaming server can be reached	""
nats.external.port	The port at which the externally-managed NATS Streaming server can be reached	""
nats.enableMonitoring	Enable the NATS monitoring endpoints on port 8222 for NATS Streaming deployments managed by this chart	false
faasIdler.create	Create the faasIdler component	true
faasIdler.inactivityDuration	Duration after which faas-idler will scale function down to 0	15m
faasIdler.reconcileInterval	The time between each of reconciliation	1m
faasIdler.dryRun	When set to false the OpenFaaS API will be called to scale down idle functions, by default this is set to only print in the logs.	true
prometheus.create	Create the Prometheus component	true
alertmanager.create	Create the AlertManager component	true
istio.mtls	Create Istio policies and destination rules to enforce mTLS for OpenFaaS components and functions	false

Specify each parameter using the --set key=value[,key=value] argument to helm install. See values.yaml for detailed configuration.