Kubernetes & microservices

Microservices vs Monolith


Kubernetes (abr. k8s) is a container orchestration system designed for highly scalable cloud applications. The main advantages of using k8s over historical server management systems are the following:

A nice quote that I could not find the source is:

Kubernetes is all the perl, bash and python scripts system administrators have put together during the 30 previous years

Quick startup using minikube

Minikube is a a tool that helps running k8s on a local machine using (by default) virtual machines to simulate the network.

Starting a minikube cluster:

minikube start

On Linux minikube can be used without a virtual machine drive (using Docker or another container runtime):

minikube start --vm-driver=none



VIPs (Virtual IP Addresses) are alocated using kube-proxy. Since 1.2 k8s uses iptables as the default proxying mode, but userland solution also exists.

The userland solution has the caveat of having more overhead (ring0-ring3 packet forwarding) but it has the advantage of beeing able to retry requests if a backend pod is not available. The later problem can be fixed using readiness probes.

The iptables proxy mode selects a backend pod randomly.

In k8s v1.11 IPVS proxy mode was implemented, it is based on netfilter hook function and so have better latency, throughput, and feature set.

IPVS documentation: https://linux.die.net/man/8/ipvsadm

Service discovery

There is two main mode for service discovery: DNS and environement variables

Environement variables are pushed by the container runtime at pod startup with the list of running services. For example if a redis-master service is running the following variables will be created:


NOTE: The services must be started before the Pod creation, otherwise the environment variables will not be populated. The problem does not exists with DNS-based service discovery.

A headless service is a service without a VIP and no load-balancing


Helm is a "package manager" for k8s that helps install new services on it such as Apache Spark, gitlab worker, prometheus...


Istio is a service mesh: manage services and connections between them.