Availability SLO 99.9% over a rolling 30 days. The page fires when the burn rate measured over a 1 hour window and over a 5 minute window both exceed 14.4, evaluated continuously. Until 14:00:00 no request fails. At 14:00:00 a config push makes exactly 10% of requests fail at a constant request rate and nobody touches anything. How many minutes after 14:00:00 does the page fire, to two decimals?

Cauê Braz
Staff DevOps & Platform Engineer
- location
- Brazil
- Currently building at
- Pareto.AI
- GitHub
- github.com/cauebraz
- linkedin.com/in/cauealvesbraz
I spend my days making sure other engineers can ship without thinking about the infrastructure underneath. Fifteen years in software, the last eleven on cloud, reliability and security. This is where I write about what that work keeps teaching me.
A query fans out to 2,000 leaf servers and answers only when the slowest leaf replies. Each leaf meets its own SLO: 1 request in 10,000 takes longer than one second, and the leaves are independent. What percentage of user requests take longer than one second, to the nearest whole percent?
The orders database is overloaded and answers every query with an error. Above it sit the backend, the frontend and the browser code, and each layer retries a failed call up to 3 times with randomized exponential backoff before giving up. How many database attempts can a single user action produce?
A worker runs with resources.limits.cpu: 200m on a node with 8 idle cores, the kubelet's default CFS period of 100 ms and the kernel's default 5 ms bandwidth slice. During a batch the process keeps 4 threads runnable at the same time, each wanting a full core, for the whole period, and the scheduler spreads them over 4 cores. In each 100 ms period, for how many milliseconds of wall clock time is the cgroup throttled?
A rollout deletes a pod with terminationGracePeriodSeconds: 30 and a preStop exec hook that runs sleep 20 to drain the load balancer. The application ignores SIGTERM and its in-flight jobs need 15 seconds to finish once draining starts. The container status shows exit code 137. How many seconds after the deletion request does the runtime send SIGKILL?
An Application Load Balancer with the default idle timeout fronts nginx targets with keepalive_timeout 5s, and the ALB reuses backend connections. Access logs show a steady trickle of 5xx lines with target_status_code "-", every one on a connection that sat idle a little over 5 seconds. Target CPU, health checks and the target 5xx metric are flat. Which HTTP status code is the ALB returning to the client?
A Route 53 hosted zone has its default SOA, ns-2048.awsdns-64.net. hostmaster.example.com. 1 7200 900 1209600 86400, and the SOA record keeps the default TTL of 900 seconds. At 14:00:00 a deploy script queried api.example.com before the A record existed, and a recursive resolver cached the NXDOMAIN. At 14:00:05 the A record was created with TTL 60. In seconds, how long after 14:00:00 can that resolver keep answering NXDOMAIN?
A single scrape target exposes a classic Prometheus histogram with 10 explicit upper bounds, from 0.005 to 10 seconds, and no +Inf bucket declared by the developer. The exposition carries only _bucket, _sum and _count. The metric has two labels, service with 4 values and route with 5, and every combination is observed. Prometheus ingests all of it with no relabeling. How many time series does this one histogram add to the TSDB?
- 2007
- First Tibia server, C and C++
- 2009
- A web design course, HTML, CSS and JavaScript
- 2011
- Software Engineer, independent
- 2014
- Software Engineer, intern then junior, Personare
- 2015
- DevOps Engineer, Personare
- 2017
- Engineering Manager, Personare
- 2019
- Staff DevOps Engineer, Personare
- 2023
- Staff DevOps Engineer, Mosaic.tech
- 2025
- Staff DevOps Engineer, Pareto AI
- 2026
- Now, still at Pareto.AI
| ID | Name | Kind | Date |
|---|---|---|---|
| CHG-0006 | What a sender does after the provider says no | technical | |
| CHG-0005 | What a Gmail spam banner doesn't say | technical | |
| CHG-0004 | When will tools start writing for agents? | personal | |
| CHG-0003 | What actually reduces histogram cardinality | technical | |
| CHG-0002 | What OpenTelemetry's Cardinality Guardian actually protects | technical | |
| CHG-0001 | The job you take to get moving | personal |
- ?
- this list
- g h
- overview
- g a
- About
- g b
- Blog
- g c
- Commitments
- g r
- Recommendations
- t
- toggle the theme
- Esc
- close