Developer blog
How we build Nazexa — architecture, trade-offs and postmortems.
How we build Nazexa: architecture, trade-offs, benchmarks and the postmortems we would rather not have written.
Recent posts
Rewriting the isolate scheduler
How we cut cold starts from 19 ms to 11 ms by changing when we pre-warm rather than how much we pre-warm.
The 41-minute EU-West degradation
A metadata cache stampede, why our circuit breaker did not fire, and the three changes we shipped afterwards.
Copy-on-write branching at 500 GB
Reproducible methodology, published harness, and an honest look at where our numbers get worse.
Consistency without ceremony
How multi-region writes pick a conflict strategy, and why we refused to make it the default.
Our writing rules
We publish the methodology alongside every benchmark, including the hardware, the dataset and the harness, so anyone can reproduce or refute the result.
Postmortems name the failure, not the person. They include the timeline, the customer impact in real units, and the specific changes with links to the pull requests that made them.
- RSS feed
- Monthly digest email
- Talks and recordings
- Open-source benchmark harness
Read the archive
Four years of engineering writing, all of it free and none of it gated behind a form.