Scale Is a Sequence, Not a Destination
Systems do not break at scale. They break in a predictable order, and knowing the order is most of the job.
Scale & Performance
Everything filed under Scale & Performance — essays, notes, playbooks and signals.
Watch read replicas stop helping
Replicas fix read saturation and nothing else. Here is the exact point where adding another one stops paying for itself.
Kafka consumers I have regretted
A catalogue of my own mistakes, kept partly as penance and partly as a checklist.
p99 is a team boundary, not a metric
Whoever owns the p99 owns the on-call pager. Everything else follows from that.
Measure before caching
A cache added without a measurement is a correctness risk you took for an unknown gain.
Outbox before dual writes
If a write has to land in two places, put it in one place and let a reader fan it out.
Read replicas before sharding
When read load is the bottleneck, the cheapest correct answer is almost never sharding.
Slow developer experience will bottleneck fast models
If the model is faster than your CI, your CI is now the product constraint.