Serverless

Introduced in GitLab 11.5.

CAUTION: Caution: Serverless is currently in alpha.

Run serverless workloads on Kubernetes using Knative.

Overview

Knative extends Kubernetes to provide a set of middleware components that are useful to build modern, source-centric, container-based applications. Knative brings some significant benefits out of the box through its main components:

Serving: Request-driven compute that can scale to zero.
Eventing: Management and delivery of events.

For more information on Knative, visit the Knative docs repo.

With GitLab Serverless, you can deploy both functions-as-a-service (FaaS) and serverless applications.

Prerequisites

To run Knative on Gitlab, you will need:

Existing GitLab project: You will need a GitLab project to associate all resources. The simplest way to get started:
- If you are planning on deploying functions, clone the functions example project to get started.
- If you are planning on deploying a serverless application, clone the sample Knative Ruby App to get started.
Kubernetes Cluster: An RBAC-enabled Kubernetes cluster is required to deploy Knative. The simplest way to get started is to add a cluster using GitLab's GKE integration. The set of minimum recommended cluster specifications to run Knative is 3 nodes, 6 vCPUs, and 22.50 GB memory.
Helm Tiller: Helm is a package manager for Kubernetes and is required to install Knative.
GitLab Runner: A runner is required to run the CI jobs that will deploy serverless applications or functions onto your cluster. You can install the GitLab Runner onto the existing Kubernetes cluster. See Installing Applications for more information.
Domain Name: Knative will provide its own load balancer using Istio. It will provide an external IP address or hostname for all the applications served by Knative. You will be prompted to enter a wildcard domain where your applications will be served. Configure your DNS server to use the external IP address or hostname for that domain.
.gitlab-ci.yml: GitLab uses Kaniko to build the application. We also use gitlabktl CLI to simplify the deployment of services and functions to Knative.
serverless.yml (for functions only): When using serverless to deploy functions, the serverless.yml file will contain the information for all the functions being hosted in the repository as well as a reference to the runtime being used.
Dockerfile (for applications only: Knative requires a Dockerfile in order to build your applications. It should be included at the root of your project's repo and expose port 8080. Dockerfile is not require if you plan to build serverless functions using our runtimes.
Prometheus (optional): Installing Prometheus allows you to monitor the scale and traffic of your serverless function/application. See Installing Applications for more information.

Installing Knative via GitLab's Kubernetes integration

NOTE: Note: The minimum recommended cluster size to run Knative is 3-nodes, 6 vCPUs, and 22.50 GB memory. RBAC must be enabled.

Add a Kubernetes cluster and install Helm.
Once Helm has been successfully installed, scroll down to the Knative app section. Enter the domain to be used with your application/functions (e.g. example.com) and click Install.
After the Knative installation has finished, you can wait for the IP address or hostname to be displayed in the Knative Endpoint field or retrieve the Istio Ingress Endpoint manually.

NOTE: Note: Running kubectl commands on your cluster requires setting up access to the cluster first. For clusters created on GKE, see GKE Cluster Access, for other platforms Install kubectl.
The ingress is now available at this address and will route incoming requests to the proper service based on the DNS name in the request. To support this, a wildcard DNS A record should be created for the desired domain name. For example, if your Knative base domain is knative.info then you need to create an A record or CNAME record with domain *.knative.info pointing the ip address or hostname of the ingress.

NOTE: Note: You can deploy either functions or serverless applications on a given project but not both. The current implementation makes use of a serverless.yml file to signal a FaaS project.

Using an existing installation of Knative

Introduced in GitLab 12.0.

NOTE: Note: The "invocations" monitoring feature of GitLab serverless will not work when adding an existing installation of Knative.

It is also possible to use GitLab Serverless with an existing Kubernetes cluster which already has Knative installed.

You must do the following:

Follow the steps to add an existing Kubernetes cluster.
Ensure GitLab can manage Knative:
- For a non-GitLab managed cluster, ensure that the service account for the token provided can manage resources in the serving.knative.dev API group.
- For a GitLab managed cluster, if you added the cluster in GitLab 12.1 or later, then GitLab will already have the required access and you can proceed to the next step.
  
  Otherwise, you need to manually grant GitLab's service account the ability to manage resources in the serving.knative.dev API group. Since every GitLab service account has the edit cluster role, the simplest way to do this is with an aggregated ClusterRole adding rules to the default edit cluster role: First, save the following YAML as knative-serving-only-role.yaml:
```
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: knative-serving-only-role
  labels:
    rbac.authorization.k8s.io/aggregate-to-edit: "true"
rules:
- apiGroups:
  - serving.knative.dev
  resources:
  - configurations
  - configurationgenerations
  - routes
  - revisions
  - revisionuids
  - autoscalers
  - services
  verbs:
  - get
  - list
  - create
  - update
  - delete
  - patch
  - watch
```
  Then run the following command:
```
kubectl apply -f knative-serving-only-role.yaml
```
  If you would rather grant permissions on a per service account basis, you can do this using a Role and RoleBinding specific to the service account and namespace.
Follow the steps to deploy functions or serverless applications onto your cluster.

Deploying functions

Introduced in GitLab 11.6.

Using functions is useful for dealing with independent events without needing to maintain a complex unified infrastructure. This allows you to focus on a single task that can be executed/scaled automatically and independently.

Currently the following runtimes are offered:

ruby
node.js
Dockerfile

Dockerfile presence is assumed when a runtime is not specified.

You can find and import all the files referenced in this doc in the functions example project.

Follow these steps to deploy a function using the Node.js runtime to your Knative instance (you can skip these steps if you've cloned the example project):

Create a directory that will house the function. In this example we will create a directory called echo at the root of the project.
Create the file that will contain the function code. In this example, our file is called echo.js and is located inside the echo directory. If your project is:
- Public, continue to the next step.
- Private, you will need to create a GitLab deploy token with gitlab-deploy-token as the name and the read_registry scope.
.gitlab-ci.yml: this defines a pipeline used to deploy your functions. It must be included at the root of your repository:
```
include:
  template: Serverless.gitlab-ci.yml

functions:build:
  extends: .serverless:build:functions
  environment: production

functions:deploy:
  extends: .serverless:deploy:functions
  environment: production
```
This .gitlab-ci.yml creates jobs that invoke some predefined commands to build and deploy your functions to your cluster.

Serverless.gitlab-ci.yml is a template that allows customization. You can either import it with include parameter and use extends to customize your jobs, or you can inline the entire template by choosing it from Apply a template dropdown when editing the .gitlab-ci.yml file through the user interface.

serverless.yml: this file contains the metadata for your functions, such as name, runtime, and environment.

It must be included at the root of your repository. The following is a sample echo function which shows the required structure for the file.

You can find the relevant files for this project in the functions example project.

service: functions
description: "GitLab Serverless functions using Knative"

provider:
  name: triggermesh
  environment:
    FOO: value

functions:
  echo-js:
    handler: echo-js
    source: ./echo-js
    runtime: https://gitlab.com/gitlab-org/serverless/runtimes/nodejs
    description: "node.js runtime function"
    environment:
      MY_FUNCTION: echo-js

Explanation of the fields used above:

`service`

Parameter	Description
`service`	Name for the Knative service which will serve the function.
`description`	A short description of the `service`.

`provider`

Parameter	Description
`name`	Indicates which provider is used to execute the `serverless.yml` file. In this case, the TriggerMesh middleware.
`environment`	Includes the environment variables to be passed as part of function execution for all functions in the file, where `FOO` is the variable name and `BAR` are he variable contents. You may replace this with you own variables.

`functions`

In the serverless.yml example above, the function name is echo and the subsequent lines contain the function attributes.

Parameter	Description
`handler`	The function's name.
`source`	Directory with sources of a functions.
`runtime` (optional)	The runtime to be used to execute the function. When the runtime is not specified, we assume that `Dockerfile` is present in the function directory specified by `source`.
`description`	A short description of the function.
`environment`	Sets an environment variable for the specific function only.

After the gitlab-ci.yml template has been added and the serverless.yml file has been created, pushing a commit to your project will result in a CI pipeline being executed which will deploy each function as a Knative service. Once the deploy stage has finished, additional details for the function will appear under Operations > Serverless.

This page contains all functions available for the project, the description for accessing the function, and, if available, the function's runtime information. The details are derived from the Knative installation inside each of the project's Kubernetes cluster. Click on each function to obtain detailed scale and invocation data.

The function details can be retrieved directly from Knative on the cluster:

kubectl -n "$KUBE_NAMESPACE" get services.serving.knative.dev

The sample function can now be triggered from any HTTP client using a simple POST call:

Using curl (replace the URL on the last line with the URL of your application):

curl \
--header "Content-Type: application/json" \
--request POST \
--data '{"GitLab":"FaaS"}' \
http://functions-echo.functions-1.functions.example.com/

Using a web-based tool (ie. postman, restlet, etc)

Deploying Serverless applications

Introduced in GitLab 11.5.

NOTE: Note: You can reference and import the sample Knative Ruby App to get started.

Add the following .gitlab-ci.yml to the root of your repository (you may skip this step if you've previously cloned the sample Knative Ruby App mentioned above):

include:
  template: Serverless.gitlab-ci.yml

build:
  extends: .serverless:build:image

deploy:
  extends: .serverless:deploy:image

Serverless.gitlab-ci.yml is a template that allows customization. You can either import it with include parameter and use extends to customize your jobs, or you can inline the entire template by choosing it from Apply a template dropdown when editing the .gitlab-ci.yml file through the user interface.

A serverless.yml file is not required when deploying serverless applications.

Deploy the application with Knative

With all the pieces in place, the next time a CI pipeline runs, the Knative application will be deployed. Navigate to CI/CD > Pipelines and click the most recent pipeline.

Obtain the URL for the Knative deployment

Go to the CI/CD > Pipelines and click on the pipeline that deployed your app. Once all the stages of the pipeline finish, click the deploy stage.

The output will look like this:

Running with gitlab-runner 12.1.0-rc1 (6da35412)
  on prm-com-gitlab-org ae3bfce3
Using Docker executor with image registry.gitlab.com/gitlab-org/gitlabktl:latest ...
Running on runner-ae3bfc-concurrent-0 via runner-ae3bfc ...
Fetching changes...
Authenticating with credentials from job payload (GitLab Registry)
$ /usr/bin/gitlabktl application deploy
Welcome to gitlabktl tool
time="2019-07-15T10:51:07Z" level=info msg="deploying registry credentials"
Creating app-hello function
Waiting for app-hello ready state
Service app-hello URL: http://app-hello.serverless.example.com
Job succeeded

The second to last line, labeled Service domain contains the URL for the deployment. Copy and paste the domain into your browser to see the app live.

Function details

Go to the Operations > Serverless page and click on one of the function rows to bring up the function details page.

The pod count will give you the number of pods running the serverless function instances on a given cluster.

Prometheus support

For the Knative function invocations to appear, Prometheus must be installed.

Once Prometheus is installed, a message may appear indicating that the metrics data is loading or is not available at this time. It will appear upon the first access of the page, but should go away after a few seconds. If the message does not disappear, then it is possible that GitLab is unable to connect to the Prometheus instance running on the cluster.

Enabling TLS for Knative services

By default, a GitLab serverless deployment will be served over http. In order to serve over https you must manually obtain and install TLS certificates.

The simplest way to accomplish this is to use Certbot to manually obtain Let's Encrypt certificates. Certbot is a free, open source software tool for automatically using Let’s Encrypt certificates on manually-administrated websites to enable HTTPS.

NOTE: Note: The instructions below relate to installing and running Certbot on a Linux server and may not work on other operating systems.

Install Certbot by running the certbot-auto wrapper script. On the command line of your server, run the following commands:

wget https://dl.eff.org/certbot-auto
sudo mv certbot-auto /usr/local/bin/certbot-auto
sudo chown root /usr/local/bin/certbot-auto
chmod 0755 /usr/local/bin/certbot-auto
/usr/local/bin/certbot-auto --help

To check the integrity of the certbot-auto script, run:

wget -N https://dl.eff.org/certbot-auto.asc
gpg2 --keyserver ipv4.pool.sks-keyservers.net --recv-key A2CFB51FA275A7286234E7B24D17C995CD9775F2
gpg2 --trusted-key 4D17C995CD9775F2 --verify certbot-auto.asc /usr/local/bin/certbot-auto

The output of the last command should look something like:

gpg: Signature made Mon 10 Jun 2019 06:24:40 PM EDT
gpg:                using RSA key A2CFB51FA275A7286234E7B24D17C995CD9775F2
gpg: key 4D17C995CD9775F2 marked as ultimately trusted
gpg: checking the trustdb
gpg: marginals needed: 3  completes needed: 1  trust model: pgp
gpg: depth: 0  valid:   1  signed:   0  trust: 0-, 0q, 0n, 0m, 0f, 1u
gpg: next trustdb check due at 2027-11-22
gpg: Good signature from "Let's Encrypt Client Team <letsencrypt-client@eff.org>" [ultimate]

Run the following command to use Certbot to request a certificate using DNS challenge during authorization:

./certbot-auto certonly --manual --preferred-challenges dns -d '*.<namespace>.example.com'

Where <namespace> is the namespace created by GitLab for your serverless project (composed of <project_name>-<project_id>-<environment>) and example.com is the domain being used for your project. If you are unsure what the namespace of your project is, navigate to the Operations > Serverless page of your project and inspect the endpoint provided for your function/app.

In the above image, the namespace for the project is node-function-11909507 and the domain is knative.info, thus certificate request line would look like this:

./certbot-auto certonly --manual --preferred-challenges dns -d '*.node-function-11909507.knative.info'

The Certbot tool walks you through the steps of validating that you own each domain that you specify by creating TXT records in those domains. After this process is complete, the output should look something like this:

IMPORTANT NOTES:
- Congratulations! Your certificate and chain have been saved at:
  /etc/letsencrypt/live/namespace.example.com/fullchain.pem
  Your key file has been saved at:
  /etc/letsencrypt/live/namespace.example/privkey.pem
  Your cert will expire on 2019-09-19. To obtain a new or tweaked
  version of this certificate in the future, simply run certbot-auto
  again. To non-interactively renew *all* of your certificates, run
  "certbot-auto renew"
-----BEGIN PRIVATE KEY-----
- Your account credentials have been saved in your Certbot
  configuration directory at /etc/letsencrypt. You should make a
  secure backup of this folder now. This configuration directory will
  also contain certificates and private keys obtained by Certbot so
  making regular backups of this folder is ideal.

Create certificate and private key files. Using the contents of the files returned by Certbot, we'll create two files in order to create the Kubernetes secret:

Run the following command to see the contents of fullchain.pem:

sudo cat /etc/letsencrypt/live/node-function-11909507.knative.info/fullchain.pem

Output should look like this:

Create a file with the name cert.pem with the contents of the entire output.

Once cert.pem is created, run the following command to see the contents of privkey.pem:

sudo cat /etc/letsencrypt/live/namespace.example/privkey.pem

Output should look like this:

Create a new file with the name cert.pk with the contents of the entire output.

Create a Kubernetes secret to hold your TLS certificate, cert.pem, and the private key cert.pk:

NOTE: Note: Running kubectl commands on your cluster requires setting up access to the cluster first. For clusters created on GKE, see GKE Cluster Access. For other platforms, install kubectl.
```
kubectl create --namespace istio-system secret tls istio-ingressgateway-certs \
--key cert.pk \
--cert cert.pem
```
Where cert.pem and cert.pk are your certificate and private key files. Note that the istio-ingressgateway-certs secret name is required.

Configure Knative to use the new secret that you created for HTTPS connections. Run the following command to open the Knative shared gateway in edit mode:

kubectl edit gateway knative-ingress-gateway --namespace knative-serving

Update the gateway to include the following tls: section and configuration:

tls:
  mode: SIMPLE
  privateKey: /etc/istio/ingressgateway-certs/tls.key
  serverCertificate: /etc/istio/ingressgateway-certs/tls.crt

Example:

apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
  # ... skipped ...
spec:
  selector:
    istio: ingressgateway
  servers:
    - hosts:
        - "*"
      port:
        name: http
        number: 80
        protocol: HTTP
    - hosts:
        - "*"
      port:
        name: https
        number: 443
        protocol: HTTPS
      tls:
        mode: SIMPLE
        privateKey: /etc/istio/ingressgateway-certs/tls.key
        serverCertificate: /etc/istio/ingressgateway-certs/tls.crt

After your changes are running on your Knative cluster, you can begin using the HTTPS protocol for secure access your deployed Knative services. In the event a mistake is made during this process and you need to update the cert, you will need to edit the gateway knative-ingress-gateway to switch back to PASSTHROUGH mode. Once corrections are made, edit the file again so the gateway will use the new certificates.