Canary Release & Traffic Splitting Simulator
Release model v2 to a slice of real traffic, give it a bug, and watch the error budget decide whether to roll forward or snap back — compared with a Blue/Green switch that hits everyone at once.
Predict first
Commit to a guess before you touch anything. Being wrong here is how it sticks.
10% of 200 QPS goes to a canary with a 2% error rate. How many errors in 10 minutes?
Experiment
Move the controls and watch the outputs. Goals tick off automatically when you reach them.
Canary release
Explain it
Put what you saw into words. This is what the exam will ask you to do.
How does a canary reduce risk compared with Blue/Green, and what should trigger an automatic rollback?
Code it
Write real code against real data. Python and SQL run right here in your browser and hidden tests check your work; the AI reviews it like an examiner.
An automated canary gate
Write canary_decision(baseline, canary, budget) returning "rollback", "hold" or "promote".
baseline and canary are dicts with requests, errors, p99_ms and high_risk_rate
(share of predictions flagged high-risk). budget has max_error_rate, max_p99_ms and
max_risk_shift (allowed absolute change in high-risk rate vs baseline).
- rollback if the canary breaches any budget.
- hold if it's within budget but has fewer than 1,000 requests (not enough evidence yet).
- promote otherwise.
Challenge
Fixed variations, a new problem every day, and unlimited AI challenges built from your current settings.
Canary Deployment Initial Traffic Allocation
A hospital network processes 8,000 diagnostic queries per day. During the first phase of a canary release, 5% of traffic is shifted to Model v2. Exactly how many patient requests are routed to the canary model on day one?