The Hipster Store (HS) is a sample project created by Google to illustrate a microservices architecture designed to run on Kubernetes (k8s). The HS makes for a great teaching application for Network Policy (NP) for a few reasons: its has lots of components with specific communication paths; its a working application; and its well documented.
This project was created to show off some best practices for labeling your application, components, routes, and network security policy (NP). Labeling is is key to both a healthy easy to manage application as well as implementing meaningful custom NP.
As part of this demo project the HS application was converted to work on the OpenShift Container Platform (OCP) and custom NP was created to only allow components talk to the other as needed to preform their job, according to the Service Architecture.
Labels allow specific infrastructure components like routes or deployments as well as workload like pods to be grouped or uniquely identified. In the HS application they are used both to group (for management) and identify for NP.
Its often useful to group things or all the things in your application for quick and easy management. For example, it a dev
environment it may be useful to delete all resources so that subsequent deployments can start with a clean slate.
To do this, add labels to each kind:
of object in your deployment manifest. In the hipster shop example the label app: hipster-store
is added.
- apiVersion: apps/v1
kind: Deployment
metadata:
name: emailservice
labels:
app: hipster-store
spec:
This allows all components to be removed with a single command:
oc delete all -l "app=hipster-shop"
The second way labels are used are to uniquely identity workload. This is done by adding labels to the template
section of the deployment manifest. These labels will be unique to running pods created from the deployment. It makes it easy to manage network communication by grouping pods, regardless of how many there are, and applying netpol
uniformly. In the HS example the role:
label is used:
template:
metadata:
labels:
app: hipster-store
role: emailservice
This results in unique identifying a component by its role within OCP as seen by the following animation:
This section will take you through assigning a role to each microservices and writing the associated network policy to allow them to communicate among themselves.
This sections is meant for to deploy-and-go. You won't learn much doing this other that have a working sample application you can review. The following steps (commands) will deploy the following components:
- Hipster Store Application
- Custom Policy
- Load Generate
Before you deploy this application you should remove any quick start
policy created automatically or otherwise. To check if you have quick start policy run the following command:
oc get netpol
The two policies allow-same-namespace
and allow-from-openshift-ingress
can be deleted while the third cannot. Its owned by a platform services bot and will be re-created if deleted.
NAME POD-SELECTOR AGE
platform-services-controlled-deny-by-default <none> 22h
allow-same-namespace 6m
allow-from-openshift-ingress 6m
Delete them with the following command:
oc delete netpol allow-same-namespace allow-from-openshift-ingress
If you ever want to re-deploy the quick start policy you can find it and the instructions here.
In order for PUs to deploy they need to be able to communicate with the k8s API. We'll go ahead and deploy our NSP first because it contains the policy to allow this to happen. We could deploy the applications first but we'll see lots of CrashLoopBackOff
messages as PU's health checks and k8s API communications fail.
oc process -f openshift/app-netpol -p NAMESPACE=$(oc project --short=true) \
| oc apply -f -
As mentioned above this allows PUs to communicate acording to the Service Architecture.
ProTip 🤓
The NAMESPACE
parameter is required because your policy will be applied to an Aporeto namespace that matches your OCP namespace. It will signal to Aporeto the scope of your policy.
Now deploy the application and components. This will spin up several pods and create a route you can use to test the application.
oc process -f openshift/app.yaml| oc apply -f -
Wait for all the pods to start; they will have a "READY" count of 1/1
as shown by the command oc get pods
. Once all pods have 1/1
in the READY state you can test the application by loading the route in your browser.
The HS comes with a sample load generator. This is great for debugging NSP because it will generate traffic across all the components. The loadgen.yaml
manifest contains its own NSP which is why the NAMESPACE
parameter is required.
oc process -f openshift/loadgen.yaml -p NAMESPACE=$(oc project --short=true) | \
oc apply -f -
You can now access your namespace via the Aporeto console and see the communication paths of the components. Green arrows mean all is well; orange mean something was not working but is now fixed; red indicates communication is failing. The direction of the arrow shows source to destination.
Use the label app=hipster-store
to cleanup all deployments, pods, routes and NSP associated with the HS in one easy step:
oc delete all,nsp -l "app=hipster-store"
This project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.
Find the original Hipster Store here on GitHub.
The Hipster Store Service Architecture, which defines the communication patterns of the application, can be found here.
Much of the application deployment manifests are taken directly from the k8s manifests supplied by Google. That source is Copyright 2018 Google LLC.