Apache Traffic Server (ATS) is a high performance, open-source, caching proxy server that is scalable and configurable. This project uses ATS as a Kubernetes(K8s) ingress
From high-level, the ingress controller talks to K8s' API and sets up watchers
on specific resources that are interesting to ATS. Then, the controller controls ATS by either(1) relay the information from K8s API to ATS, or (2) configure ATS directly.
- Alpine 3.12
- Apache Traffic Server 8.0.8
- LuaJIT 2.0.4
- Lua 5.1.4
- Go 1.12.8
- Other Packages
- luasocket 3.0rc1
- redis-lua 2.0.4
- Docker
- Kubernetes 1.18 (Minikube 1.11)
To install Docker, visit its official page and install the correct version for your system.
The walkthrough uses Minikube to guide you through the setup process. Visit the official Minikube page to install Minikube.
If you are cloning this project for development, visit Setting up Go-Lang for detailed guide on how to develop projects in Go.
For other purposes, you can use git clone
or directly download repository to your computer.
Once you have cloned the project repo and started Docker and Minikube, in the terminal:
$ eval $(minikube docker-env)
- To understand why we do this, please read Use local images by re-using the docker daemon
$ cd trafficserver-ingress-controller
$ git submodule update --init
$ docker build -t ats_alpine .
$ docker build -t tsexporter k8s/backend/trafficserver_exporter/
$ docker build -t node-app-1 k8s/backend/node-app-1/
$ docker build -t node-app-2 k8s/backend/node-app-2/
$ docker pull fluent/fluentd:v1.6-debian-1
- At this point, we have created necessary images for our example. Let's talk about what each step does:
- Step 4 builds an image to create a Docker container that will contain the Apache Traffic Server (ATS) itself, the kubernetes ingress controller, along with other software required for the controller to do its job.
- Step 5 builds an image for the trafficserver exporter. This exports the ATS statistics over HTTP for Prometheus to read.
- Steps 6 and 7 build 2 images that will serve as backends to kubernetes services which we will shortly create
$ kubectl create namespace trafficserver-test
- Create a namespace for ATS pod
$ openssl req -x509 -sha256 -nodes -days 365 -newkey rsa:2048 -keyout tls.key -out tls.crt -subj "/CN=atssvc/O=atssvc"
- Create a self-signed certificate
$ kubectl create secret tls tls-secret --key tls.key --cert tls.crt -n trafficserver-test --dry-run=client -o yaml | kubectl apply -f -
- Create a secret in the namespace just created
$ kubectl apply -f k8s/configmaps/fluentd-configmap.yaml
- Create config map for fluentd
$ kubectl apply -f k8s/traffic-server/
- will define a new kubernetes namespace named
trafficserver-test
and deploy a single ATS pod to said namespace. The ATS pod is also where the ingress controller lives.
- will define a new kubernetes namespace named
The following steps can be executed in any order, thus list numbers are not used.
-
$ kubectl apply -f k8s/apps/
- creates namespaces
trafficserver-test-2
andtrafficserver-test-3
if not already exist - creates kubernetes services and deployments for
appsvc1
andappsvc2
- deploy 2 of each
appsvc1
, andappsvc2
pods intrafficserver-test-2
, totally 4 pods in said namespace. - similarly, deploy 2 of each
appsvc1
, andappsvc2
pods intrafficserver-test-3
, totally 4 pods in this namespace. We now have 8 pods in total for the 2 services we have created and deployed in the 2 namespaces.
- creates namespaces
-
$ kubectl apply -f k8s/ingresses/
- creates namespaces
trafficserver-test-2
andtrafficserver-test-3
if not already exist - defines an ingress resource in both
trafficserver-test-2
andtrafficserver-test-3
- the ingress resource in
trafficserver-test-2
defines domain nametest.media.com
with/app1
and/app2
as its paths - both ingress resources define domain name
test.edge.com
; however,test.edge.com/app1
is only defined intrafficserver-test-2
andtest.edge.com/app2
is only defined intrafficserver-test-3
- Addtionally, an ingress resources defines HTTPS access for
test.edge.com/app2
in namespacetrafficserver-test-3
- creates namespaces
When both steps above have executed at least once, ATS proxying will have started to work. To see proxy in action, we can use curl:
$ curl -vH "HOST:test.media.com" "$(minikube ip):30000/app1"
$ curl -vH "HOST:test.media.com" "$(minikube ip):30000/app2"
$ curl -vH "HOST:test.edge.com" "$(minikube ip):30000/app1"
$ curl -vH "HOST:test.edge.com" "$(minikube ip):30000/app2"
$ curl -vH "HOST:test.edge.com" -k "https://$(minikube ip):30043/app2"
Below is an example of configuring Apache Traffic Server reloadable configurations using kubernetes configmap resource:
$ kubectl apply -f k8s/configmaps/
- create a ConfigMap resource in
trafficserver-test
if not already exist - configure 3 reloadable ATS configurations:
proxy.config.output.logfile.rolling_enabled: "1"
proxy.config.output.logfile.rolling_interval_sec: "3000"
proxy.config.restart.active_client_threshold: "0"
- create a ConfigMap resource in
You can attach ATS lua script to an ingress object and ATS will execute it for requests matching the routing rules defined in the ingress object. See an example in annotation section of yaml file here
You can provide an environment variable called INGRESS_CLASS
in the deployment to specify the ingress class. Only ingress object with annotation kubernetes.io/ingress.class
with value equal to the environment variable value will be used by ATS for routing
This project ships with Fluentd already integrated with the Apache Traffic Server. The configuration file used for the same can be found here
As can be seen from the default configuration file, Fluentd reads the Apache Traffic Server access logs located at /usr/local/var/log/trafficserver/squid.log
and outputs them to stdout
. The ouput plugin for Fluentd can be changed to send the logs to any desired location supported by Fluentd including Elasticsearch, Kafka, MongoDB etc. You can read more about output plugins here.
Use the following steps to install Prometheus and Grafana and use them to monitor the Apache Traffic Server statistics.
$ kubectl apply -f k8s/prometheus/ats-stats.yaml
- Creates a new service which connects to the ATS pod on port 9122. This service will be used by Prometheus to read the Apache Traffic Server stats.
$ kubectl apply -f k8s/configmaps/prometheus-configmap.yaml
- Creates a new configmap which holds the configuration file for Prometheus. You can modify this configuration file to suit your needs. More about that can be read here
$ kubectl apply -f k8s/prometheus/prometheus-deployment.yaml
- Creates a new deployment consisting of Prometheus and Grafana. Also creates two new services to access prometheus and grafana.
- Open
x.x.x.x:30090
in your web browser to access Prometheus wherex.x.x.x
is the IP returned by the command:$ minikube ip
- Open
x.x.x.x:30030
in your web browser to access the Grafana dashboard wherex.x.x.x
is the IP returned by the command:$ minikube ip
. - The default credentials for logging into Grafana are
admin:admin
- Click on `Add your first data source' and select Prometheus under the 'Time series databases category'
- Set an appropriate name for the data source and enter
localhost:9090
as the URL - Click on 'Save & Test'. If everything has been installed correctly you should get a notification saying 'Data source is working'
- Click on the '+' icon in the left handside column and select 'Dashboard'
- Click on '+ Add new panel'
- Enter a PromQL query. For example if you want to add a graph showing the total number of responses over time enter
trafficserver_responses_total
and press Shift + Enter. - Click on Apply to add the graph to your dashboard. You can similarly make add more graphs to your dashboard to suit your needs. To learn more about Grafana click here
- Get Go-lang 1.12 from official site
- Add
go
command to your PATH:export PATH=$PATH:/usr/local/go/bin
- Define GOPATH:
export GOPATH=$(go env GOPATH)
- Add Go workspace to your PATH:
export PATH=$PATH:$(go env GOPATH)/bin
- Define Go import Paths
- Go's import path is different from other languages in that all import paths are absolute paths. Due to this reason, it is important to set up your project paths correctly
- define the base path:
mkdir -p $GOPATH/src/github.com/
- Clone the project:
cd $GOPATH/src/github.com/
git clone <project>
- As of Go 1.12 in order to have
go.mod
within Go paths, you must export:export GO111MODULE=on
to be able to compile locally.
To compile, type: go build -o ingress_ats main/main.go
The project includes unit tests for the controller written in Golang and the plugin written in Lua.
To run the Golang unit tests: go test ./watcher/ && go test ./redis/
The Lua unit tests use busted
for testing. busted
can be installed using luarocks
:luarocks install busted
. More information on how to install busted is available here.
โ ๏ธ Note that the project uses Lua 5.1 version
To run the Lua unit tests:
cd pluginats
busted connect_redis_test.lua
The repository comes with basic support for both vscode and vim
.
If you're using vscode
:
.vscode/settings.json
contains some basic settings for whitespaces and tabs.vscode/extensions.json
contains a few recommended extensions for this project. It is highly recommended to install the Go extension since it contains the code lint this project used during development.
If you're using vim
, a vimrc
file with basic whitespace and tab configurations is also provided