Kafka in anger: what I still use from telco work
Years around cloud-native telecom stacks left me with opinions about topics, partitions, and operational honesty—not buzzwords. What transferred when I left the Swedish campus for Athens.
I did a long stretch in cloud-native telco work—Kubernetes, Kafka, security reviews, the kind of systems where “we will fix it in post” is not an option. Some of it was vendor-specific magic. A smaller slice turned out to be portable habits I still reach for on marketplace and streaming projects.
Topics are not free jazz
Kafka is happy to let you create chaos. I care about topic naming, retention that matches legal reality, and consumer groups that fail loudly when lag crosses a threshold you can defend in a meeting. If nobody can explain why a topic exists, it will become a dumping ground.
Backpressure is a user experience issue
When downstream slows, the answer is not always “more partitions.” Sometimes it is admission control, shedding non-critical traffic, or admitting that your consumer is doing too much per message. I try to pair every streaming path with a number someone agreed to: max lag, max batch time, max retry budget.
Security reviews are a feature
The parts that felt bureaucratic at 22:00 were often the ones that stopped a bad credential scope from shipping. I carry that instinct into smaller teams: least privilege by default, short-lived credentials, and secrets that never get copy-pasted into Slack “just for debugging.”
What I deliberately left behind
Not every pattern from massive org charts fits a two-person sprint. I do not import six layers of approval to merge a hotfix—but I do import blameless postmortems and runbooks that a tired human can follow.
Infrastructure work is still people work. The best systems are the ones where the next engineer mutters “thank you” under their breath while reading the README.
If you are designing eventing or platform boundaries and want a skeptical read, you know where to find me on this site.