Skip to selected work
Engineering portfolioParis area · France

Rayan
Es-Seffar

DevOps & Cloud Engineer

I run Kubernetes workloads on AWS for advertising and analytics platforms. I built the HAProxy setup that routes traffic into our Kubernetes cluster and migrated our applications from Traefik.

Explore my experience
Explore the project hubs

Experience

DevOps Full-Stack Developer

YsoCorpMarch 2025 – Present

Maintained AWS infrastructure for advertising and analytics platforms, from Kubernetes rollouts to application logging.

Backend and DevOps Developer

FreelanceAugust 2024 – Present

Worked directly with clients to design, deploy, and maintain their infrastructure.

Selected projects

Replacing Traefik with HAProxy on AWS

AWS · Kubernetes · HAProxy · Terraform

Building the replacement

I set up HAProxy on Kubernetes, integrated it with the AWS Network Load Balancer, and configured the routes that sent incoming requests to the right applications.

Moving traffic

I pointed the cluster subdomain at HAProxy through Route 53. Once HAProxy was serving the applications, I removed the old Traefik routes, middleware, monitoring, and dedicated node-pool configuration from the infrastructure repository.

Operating HAProxy under load

I investigated health checks, reload behavior, and uneven traffic across pods. When replica counts repeatedly rose and fell, I raised the minimum capacity and configured the autoscaler to wait longer before scaling down.

Discuss this project

I built the HAProxy ingress setup, moved application traffic from Traefik, retired the old configuration, and tuned the replacement under production traffic.

Finding a CPU hotspot in a live Rust API

Linux perf · Flamegraphs · Kubernetes

Where was the CPU time going?

The API handled real requests with different payloads. I wanted to see which parts of the request path consumed CPU under that mix of traffic.

How I measured it

I used a separate Kubernetes pod with profiling symbols, briefly routed live requests to it, and captured a flamegraph. After the capture, I detached the pod and checked that it had left the service.

What the profile showed

Across roughly 40,000 requests, the compression routine accounted for about 20% of sampled CPU event weight. That gave me a specific operation to investigate before changing the service.

Discuss this project

Across roughly 40,000 live requests, I traced about 20% of sampled CPU work to one compression routine.

Building a Druid connector for Apache Kudu

Apache Druid · Apache Kudu · Java · Docker · CI

Connecting the systems

The analytics engine needed a way to read records from a separate data store. A connector also had to reach the runtime environment alongside the engine.

What I built

I implemented a Kudu input source for Druid, added filters for selecting records and wired a custom image build into CI.

Keeping it usable

When the engine and Java runtime were upgraded, I updated the extension and image build with them. The connector became part of the platform’s release process.

Discuss this project

I built a Java connector that lets Druid read filtered records from Kudu and packaged it into the platform’s release pipeline.

Contact Rayan