Jonas Petrik

The business needed checkout to handle baskets with up to 500 unique items. Each item is an individually configured variant with many attributes, not a quantity of one product.

I ran more than 15 measured load-test rounds with Blackfire and Datadog. Each round found the next bottleneck. I wrote a report for each round.

I replaced per-item database queries with batched queries and lookup maps. I cached slow external calls when the input did not change.

I made logging cost flat in basket size and tuned PHP opcache with preload.

These items now work in the checkout hot path under load. A checkout with 500 unique items completed in load tests. One round alone cut a 500-item checkout time by 44%.

Checkout had to handle 1,000 checkouts per minute. Resource-based autoscaling reacted too late when all PHP workers were busy.

I replaced resource-based autoscaling with traffic-based autoscaling, using requests in flight with KEDA. I fixed readiness and liveness probes and capped in-flight work per pod.

I ran load tests on a demo environment and in production, with no real orders.

Traffic-based autoscaling is now live in production and replaced the old setup.

The production load test held 1,000 checkouts per minute for 10 minutes, with 0 dropped. 99.97% of checkouts completed, and the service scaled out about 10× on its own. No pod was killed or evicted.

I took ownership of on-call quality for checkout. Noisy alerts paged engineers for the wrong reasons.

I fixed alert quality and the failure paths behind the noisy alerts. I wrote the team's monitoring-duty process.

I set up Blackfire profiling for three PHP runtimes (RoadRunner, FrankenPHP, PHP-FPM), with linked profiles across services. I fixed gaps in Datadog APM traces.

I audited all checkout SLOs against what Datadog really measured. I wrote 15+ incident reports with fixes.

On-call had less alert noise, and engineers trusted the alerts more. This work was part of the reason for my promotion to the lead role.

The old checkout was hard to change and slow under load.

I was one of the key engineers who built the new checkout API on Laravel and RoadRunner, and I led many of its design decisions.

In 2024 it replaced the old checkout for every brand on the platform. I then ran the Black Friday 2024 load test, and checkout passed well above its goal.

Later, as lead, I designed the plan to separate checkout from the legacy order system, so that its outages will not stop customers from paying.

anwalt.de is Germany's largest legal marketplace, with 4M+ monthly users. It needed stronger search and better foundations for product work.

I rebuilt search on Elasticsearch with GraphQL, geolocation, aggregations and range filters.

I then led a 3-person team that built delivery and platform foundations. Other teams adopted them.

The rebuild led to my promotion to the lead role, and my work expanded to CI/CD, AWS infrastructure and mentoring.

Atobi is a retail employee platform for brands like Nike and PVH Nordic. It started on a prototype stack.

I led the migration to a Laravel API, a React frontend and mobile apps.

I designed a reusable gamification system for engagement and calls to action. Multiple products used it.

The team grew from 3 to 10 engineers while I stayed hands-on with architecture.