Explain an incident without hindsight
Reconstruct the response using what people knew at the time. Explain contributing conditions and decision constraints rather than assigning a convenient single cause.
Articles
Guides, news and explanations about AI and reliable systems.
Not sure where to begin? Choose a short reading path, or browse at your own pace.
Recover a broken Linux service from evidence
Break one part of a disposable Linux service, follow the evidence and restore the original page. A 200 response is only the start of the investigation.
Collect the first signals from a local service
Create one synthetic export and find the same job in a reply, log events, counters and a local trace Collector.
Write logs that let you follow one job
Give each event enough context to find the right export, distinguish a retry from a new job, and see where the recorded story still has a gap.
Test an alert from condition to responder
A correct alert condition can still reach nobody. Test the calculation, delivery, message links and missing-data behavior as separate parts of one journey.
Follow a trace through your local OpenTelemetry Collector
Add a batch processor to the active pipeline, make a broken reference visible, and see what a local delivery test can tell you about the collector.
Write a page the next responder can use
A delayed export needs a useful first investigation. Connect the interruption to customer impact, a responsible person and a way to recognize recovery.
Why a Prometheus chart can show work going backward
A restart resets a counter, not the work already completed. Follow the query into rates, missing scrapes and the difference between an average wait and a slow tail.
Find out what your canary release actually tested
Choose work that reaches the changed code, compare like with like, and plan what happens to files and unfinished jobs if you stop the rollout.
Complete an investigation another engineer can repeat
Compare two controlled Dispatch runs, test a competing explanation and repair broken trace context. Leave commands and evidence the next engineer can follow.
Follow a request into its background job
Break and restore the link between a quick API response and the slower export it starts, while keeping the job itself separate from its recorded trace.
Decide how SRE, DevOps and platform teams share the work
Compare service ownership, shared platforms and SRE engagement; use the reliability plan to decide where the work belongs.