The Zero Config MicroService using Kubernetes and Azure KeyVault

Mom said don’t talk to strangers, in the new world Mom said never share your secret key!

TL;DR: In this blog post, we demonstrate the value of Centralised configuration and secret stores by leveraging Kubernetes and Azure KeyVault for an ASP.NET Core microservice. It allows the team to create a toolchain that developers and ops engineers can use to entirely avoid creation and management of configuration files (appSettings.json, blah.xml) and focus more on the actual application development.

If you are more interested in the source code and how to setup all pieces, you can jump directly to the GitHub repo here( “The Zero Config Microservice”).

The sample provides the following capabilities:

  • A sample microservice project to demonstrate the use of Azure KeyVault and Kubernetes ConfigMaps for Configuration, it stresses on the importance of separating DevOps functions from Developer functions by having no appSettings, secret.json, blah.json Files in code. All data is injected through Kubernetes or asked specifically from Azure Key Vault.

Other features in the sample:

  • use of Serilogfor structured logging,
  • use of the repository for Azure CosmosDB, a generic repository that can be used and extended for CosmosDB Document operations.
  • deployment of core microservice container to Kubernetes

The nightmare begins…

It’s 2:00 AM, Adam is done making all changes to his super awesome code piece, the tests are all running fine, he hit commit -> push -> all commits pushed successfully to git. Happily, he drives back home. Ten mins later he gets a call from the SecurityOps engineer, – Adam did you push the Secret Key to our public repo?

YIKES!! that damn blah.config file Adam thinks, how could I have forgotten to include that in .gitignore, the nightmare has already begun ….

We can surely try to blame Adam here for committing the sin of checking in sensitive secrets and not following the recommended practices of managing configuration files, but the bigger question is that if the underlying toolchain had abstracted out the configuration management from the developer, this fiasco would have never happened!

The virus was injected a long time ago…

Since the early days of .NET, there has been the notion of app.config and web.config files which provide a playground for developers to make their code flexible by moving common configuration into these files. When used effectively, these files are proven to be worthy of dynamic configuration changes. However a lot of time we see the misuse of what goes into these files. A common culprit is how samples and documentation have been written, most samples out in the web would usually leverage these config files for storing key elements such as ConnectionStrings, and even password. The values might be obfuscated but what we are telling developers is that “hey, this is a great place to push your secrets!”. So, in a world where we are preaching using configuration files, we can’t blame the developer for not managing the governance of it. Don’t get me wrong; I am not challenging the use of Configuration here, it is an absolute need of any good implementation, I am instead debating the use of multiple json, XML, yaml files in maintaining configuration settings. Configs are great for ensuring the flexibility of the application, config files, however, in my opinion, are a pain to manage especially across environments and soon you end up in Config Hell.

A ray of hope: The DevOps movement

In recent years, we have seen a shift around following some great practices around effective DevOps and some great tools (Chef, Puppet) for managing Configuration for different languages. While these have helped to inject values during CI/CD pipeline and greatly simplified the management of configuration, the blah.config concept has not completely moved away. Frameworks like ASP.NET Core support the notion of appSettings.json across environments, the framework has made it very effective to use these across environments through interfaces like IHostingEnvironment and IConfiguration but we can do better.

Clean code: Separation of Concerns

One of the key reasons we would want to move the configuration away from source control is to delineate responsibilities. Let’s define some roles to elaborate this, none of these are new concepts but rather a high-level summary:

  • Configuration Custodian: Responsible for generating and maintaining the life cycle of configuration values, these include CRUD on keys, ensuring the security of secrets, regeneration of keys and tokens, defining configuration settings such as Log levels for each environment. This role can be owned by operation engineers and security engineering while injecting configuration files through proper DevOps processes and CI/CD implementation. Note that they do not define the actual configuration but are custodians of their management.
  • Configuration Consumer: Responsible for defining the schema (loose term) for the configuration that needs to be in place and then consuming the configuration values in the application or library code. This is the Dev. And Test teams, they should not be concerned about what the value of keys are rather what the capability of the key is, for example, a developer may need different ConnectionString in the application but does not need to know the actual value across different environments.
  • Configuration Store: The underlying store that is leveraged to store the configuration, while this can be a simple file, but in a distributed application, this needs to be a reliable store that can work across environments. The store is responsible for persisting values that modify the behavior of the application per environment but are not sensitive and does not require any encryption or HSM modules.
  • Secret Store: While you can store configuration and secrets together, it violates our separation of concern principle, so the recommendation is to leverage a separate store for persisting secrets. This allows a secure channel for sensitive configuration data such as ConnectionStrings, enables the operations team to have Credentials, Certificate, Token in one repository and minimizes the security risk in case the Configuration Store gets compromised.

The below diagram shows how these roles play together in a DevOps Inner loop and Outer loop. The inner loop is focussed on the developer teams iterating over their solution development; they consume the configuration published by the outer loop. The Ops Engineer govern the Configuration management and push changes into Azure KeyVault and Kubernetes that are further isolated per environment.

Kubernetes and Azure KeyVault to the rescue

So we have talked a lot about governance elements and the need to move out of configuration files, lets now use some of the magic available in Kubernetes and Azure Key Vault to implement this. In particular, we would be using the following capabilities of these technologies:

Why not just use Kubernetes you may ask?

That’s a valid question since Kubernetes supports both a ConfigMap store and Secret store. Remember our principle around the separation of concerns and how we can ensure enhanced security by separating out configuration from secrets. Having secrets in the same cluster as the configuration store can make them prone to higher risks. An additional benefit is Maintainability. Azure KeyVault gives the ability to provide a distributed “Secure Store as a Service” option that provides not just secret management but also Certificates and Key management as part of the service. The SecOps engineers can lock down access to the store without need of cluster admins permissions which allows for clear delineation of responsibilities and access.

PS: There is a discussion going on in Kubernetes groups to provide under the hood support for saving Kubernetes secrets directly in Azure KeyVault, “when” and “if” that happens we will have the best of both worlds implemented using the abstractions of Kubernetes, fingers crossed!

Let’s get to our scenario now.

The Scenario

We will be building a Vehicle microservice which provides CRUD operations for sending vehicle data to a CosmosDB document store. The sample micro-service needs to interact with the Configuration stores to get values such as connectionstring, database name, collection name, etc. We interact with Azure KeyVault for this purpose. Additionally, the application needs the Authentication token for Azure Key Vault itself, these details along with other Configuration will be stored in Kubernetes.

If you are more interested in the source code and how to setup all pieces, you can jump directly to the GitHub repo here.

Mapping the above diagram to our roles and responsibilities earlier:

  • The Ops engineer/scripts are the Configuration Custodian and they are the only ones who work in the outer loop to manage all the configuration. They would have CI/CD scripts that would inject these configurations or use popular framework tools to enable the insertion during the build process.
  • The Vehicle API is the ASP.NET Core 2.0 application and is the Configuration Consumer here; the consumer is interested in getting the values without really worrying about what the value is and which environment it belongs to. The ASP.NET Core framework provides excellent support for this through its Configuration extensibility support. You can add as many providers as you like and they can be bound an IConfiguration object which provides access to all the configuration. In the below code snippet, we provide the configuration to be picked up from environment variables instead of a configuration file. The ASP.NET Core 2.0 framework also supports extensions to include Azure KeyVault as a configuration provider, and under the hood, the Azure KeyVault client allows for secure access to the values required by the application.


 // add the environment variables to config

 // add azure key vault configuration using environment variables
 var buildConfig = config.Build();

 // get the key vault  uri
 var vaultUri = buildConfig["kvuri"].Replace("{vault-name}",   buildConfig["vault"]);
 // setup KeyVault store for getting configuration values
 config.AddAzureKeyVault(vaultUri, buildConfig["clientId"], buildConfig["clientSecret"]);
  • AzureKeyVault is the Secret Store for all the secrets that are application specific. It allows for the creation of these secrets and also managing the lifecycle of them. It is recommended that you have a separate Azure KeyVault per environment to ensure isolation. The following command can be used to add a new configuration into KeyVault:

#Get a list of existing secrets
az keyvault secret list --vault-name  -o table  

#add a new secret to keyvault
az keyvault secret set -n MySecret --value MyValue --description "my custom value" --vault-name 
  • Kubernetes is the Configuration Store and serves multiple purposes here:
  1. Creation of the ConfigMaps and Secret: Since we are injecting the Azure KeyVault authentication information using Kubernetes, the Ops engineer will provide these values using two constructs provided in the Kubernetes infrastructure. ConfigMaps and Secrets. The following shows how to add config maps and secrets from the Kubernetes command line:
kubectl create configmap vault --from-literal=vault=   
kubectl create configmap kvuri --from-literal=kvuri=https://{vault-name}
kubectl create configmap clientid --from-literal=clientId= 
kubectl create secret generic clientsecret --from-literal=clientSecret=

The clientsecret is the only piece of secure information we store in Kubernetes, all the application specific secrets are stored in Azure KeyVault. This is comparatively safer since the above scripts do not need to go in the same git repo. (so we don’t check them in by mistake) and can be managed separately. We still control the expiry of this secret using Azure KeyVault, so the Security engineer still has full control over access and permissions.

  1. Injecting Values into the Container: During runtime, Kubernetes will automatically push the above values as environment variables for the deployed containers, so the system does not need to worry about loading them from a configuration file. The Kubernetes configuration for the deployment looks like below. As you would notice, we only provide a reference to the ConfigMaps and Secret that have been created instead of punching in the actual values.
apiVersion: apps/v1beta1
kind: Deployment
  name: vehicle-api-deploy #name for the deployment
    app: vehicle-api #label that will be used to map the service, this tag is very important
  replicas: 1
      app: vehicle-api #label that will be used to map the service, this tag is very important
        app: vehicle-api #label that will be used to map the service, this tag is very important
      - name: vehicleapi #name for the container configuration
        image: <yourdockerhub>/<youdockerimage>:<youdockertagversion> # **CHANGE THIS: the tag for the container to be deployed
        imagePullPolicy: Always #getting latest image on each deployment
        - containerPort: 80 #map to port 80 in the docker container
        env: #set environment variables for the docker container using configMaps and Secret Keys
        - name: clientId
              name: clientid
              key: clientId
        - name: kvuri
              name: kvuri
              key: kvuri
        - name: vault
              name: vault
              key: vault
        - name: clientsecret
              name: clientsecret
              key: clientSecret
      imagePullSecrets: #secret to get details of private repo, disable this if using public docker repo
      - name: regsecret


There we go! We now have a micro-service that is driven by the powers of DevOps and eliminates the need for any configuration files. Do check out the source code in the Github repo and share your feedback.

happy coding 🙂 …