- How does it work?
- Architecture of the entire development system
In order to write software for and with Kubernetes, obviously a Kubernetes cluster is required. There are already a number of Kubernetes distributions available to run everything locally. A cloud-based Kubernetes cluster can be connected as well in order to spare the development computer from blasting off. A working KUBECONFIG connection is required with appropriate permissions which should always be the case for local clusters. Gefyra installs the required cluster-side components by itself once a development setup is about to be established.
With these components, Gefyra is able to control a local development machine, and the development cluster, too. Both sides are now in the hand of Gefyra.
Once the developer’s work is done, Gefyra well and truly removes all components from the cluster without leaving a trace.
A few things are required in order to achieve this:
- a tunnel between the local development machine and the Kubernetes cluster
- a local end of that tunnel to steer the traffic, DNS, and encrypt everything passing over the line
- a cluster end of the tunnel, forwarding traffic, taking care of the encryption
- a local DNS resolver that behaves like the cluster DNS
- sophisticated IP routing mechanisms
- a traffic interceptor for containers already running withing the Kubernetes cluster
Gefyra builds on top of the following popular open-source technologies:
Docker is currently used in order to manage the local container-based development setup, including the host, networking and container management procedures.
Wireguard is used to establish the connection tunnel between the two ends. It securely encrypts the UDP-based traffic and allows to create a site-to-site network for Gefyra. That way, the development setup becomes part of the cluster and containers running locally are actually able to reach cluster-based resources, such as databases, other (micro)services and so on.
CoreDNS provides local DNS functionality. It allows resolving resources running within the Kubernetes cluster.
Nginx is used for all kinds of proxying and reverse-proxying traffic, including the interceptions of already running containers in the cluster.
Rsync is used to synchronize directories from containers running in the cluster to local instances. This is particularly important for Kubernetes service account tokens during a bridge operation.
The local development happens with a running container instance of the application in question on the developer machine. Gefyra takes care of the local Docker host setup, and hence needs access to it. It creates a dedicated Docker network which the container is deployed to. Next to the developed application, Gefyra places a sidecar container. This container, as a component of Gefyra, is called Cargo.
Cargo acts as a network gateway for the app container and, as such, takes care of the IP routing into and from the cluster. In addition, Cargo provides a CoreDNS server which forwards all requests to the cluster. That way, the app container will be able to resolve cluster resources and may not resolve domain names that are not supposed to be resolved (think of isolated application scenarios). Cargo encrypts all the passing traffic with Wireguard using ad-hoc connection secrets.
This local setup allows developers to use their existing tooling, including their favorite code editor and debuggers. The application, when it is supported, can perform code-hot-reloading upon changes and pipe logging output to a local shell (or other systems).
Of course, developers are able to mount local storage volumes into the container, override environment variables and modify everything as they’d like to.
Replacing a container in the cluster with a local instance is called bridge: from an architectural perspective the local application is bridged into the cluster. If the container is already running within a Kubernetes Pod, it gets replaced and all traffic to the originally running container is proxied to the one on the developer machine.
During the container startup of the application, Gefyra modifies the container’s networking from the outside and sets the default gateway to Cargo. That way, all container’s traffic is passed to the cluster via Cargo’s encrypted tunnel. The same procedure can be applied for multiple app containers at the same time.
The neat part is that with a debugger and two or more bridged containers, developers can introspect requests from the source to the target and back around while being attached to both ends.
This chapter covers the important bridge operation by following an example.
Think of a provisioned Kubernetes cluster running some workload. There is an Ingress, Kubernetes Services and Pods running containers. Some of them use the sidecar (https://medium.com/nerd-for-tech/microservice-design-pattern-sidecar-sidekick-pattern-dbcea9bed783) pattern.
Before the brigde can happen, Gefyra installs all required components to the cluster. A valid and privileged connection must be available on the developer machine to do so.
The main component is the cluster agent called Stowaway. The Stowaway controls the cluster side of the tunnel connection. It is operated by Gefyra’s Operator application.
Stowaway boots up and dynamically creates Wireguard connection secrets (private/public key-pair) for itself and Cargo. Gefyra copies these secrets to Cargo for it to establish a connection. This is a UDP connection. It requires a Kubernetes Service of kind nodeport to allow the traffic to pass through for the time of an active development session. Gefyra’s operator installs these components with the requested parameters and removes it after the session terminates.
By the way: Gefyra’s operator removes all components and itself from the cluster in case the connection was disrupted for some time, too.
Once a connection could be establised from Cargo to Stowaway (after running
gefyra up), Gefyra can spin up the app container on the local side for the developer to start working (by running
gefyra run ...).
Another job of Gefyra’s operator is to rewrite the target pods, i.e. exchange the running container through Gefyras proxy, called Carrier (upon running
gefyra bridge ...).
For that, it creates a temporary Kubernetes Service that channels the Ingress traffic (or any other kind of cluster internal traffic) to the container through Stowaway and Cargo to the locally running app container.
A bridge can robustly run as long as it is required to (given the connection does not drop in the meanwhile). Looking at the example, Carrier was installed in pod <C> on port XY. That port was previously occupied by the container running there originally. In most cases, the local app container represents the development version of that originally provisioned container. Traffic coming from the Ingress, passing on to the Service <C> hits Carrier (the proxy). Carrier bends the request to flow through Gefyras Service to the local app container via Stowaway and Cargo’s tunnel. This works since the app container’s IP is routable from within the cluster.
The local app container does not simply return a response, but fires up another subsequent request by itself to Service <A>. The request roams from the local app container back into the cluster and hits Pod <A>’s container via Service <A>. The response is awaited.
Once the local app container is done with constructing its initial answer the response gets back to Carrier and afterwards to the Ingress and back to the client.
With that, the local development container is reachable exactly the same way another container from within the cluster would be. That fact is a major advantage, especially for frontend applications or domain-sensitive services.
Developers now can run local integration tests with new software while having access to all interdependent services.
Once the development job is done, Gefyra properly removes everything, resets Pod <C> to its original configuration, and tears the local environment down (just like nothing ever happened).