Oaktown Tech / practical labs
The route came back. Did the service?
Put fault detection, route changes, and actual observations side by side. Learn what the timestamps prove and what they leave unknown.
Read the text guide: instructions and limitations without interactive controls.
Controlled experiment
Same recovery. Different sampling.
In this invented experiment, the application fails at 10 seconds and recovers at 15.4 seconds. Change only when it is sampled. Predict the recovery bracket before generating the observations.
Try 5-second samples, then change the offset. Does the first passing observation move?
This is a declared synthetic truth for learning. Ordinary uploaded observations do not reveal the behavior between samples. Generating replaces the workbench CSV and declared event fields.
Experiment workbench
Return to the challenges ↑Choose the evidence
Same fault. Different conclusions.
A route change is an event. A working application is an observation. Put them on the same clock before deciding what recovered.
Inspect or edit the sample CSV
Use synthetic or non-sensitive observations. Nothing is uploaded. Keep rows time-ordered; use at most 2,000 samples and 12 series.
These are supplied samples, not live probes. No router is contacted. The declared fault is not proof of the cause, and a passing sample cannot describe the time between samples.
Read the difference
Detection is not recovery.
↑ Change the evidenceTry it before reading the answer
- In the first scenario, compare the route-change delay with the first successful probe sample.
- Choose “Ping works; the application still fails.” Which measurement would you show the customer?
- Choose the gap with no observed failure. Can you honestly call it a six-second outage?
- Choose the repeated failures. Does the first success deserve an incident-closed message?
What the evidence supports
The first scenario shows a route change 0.7 seconds after the declared fault and the first successful probe 6 seconds after it. That probe does not certify an application. In the two-series scenario, the first successful application transaction arrives after 10 seconds. Each failure-to-success transition is bracketed by the last failing sample and the next passing sample; its exact moment is unknown. Separate series do not certify each other, gaps do not establish outages, and a later failure still matters after an earlier success.
Context: RFC 5880: BFD detection. This exercise is not a certification of equipment, construction, security, or service performance.