Explain like I'm five · SRE

What is Canary Deployment?

Canary deployment, told as a story about quietly testing a new ride on just 1% of guests, in a corner.

new version, out for everyoneevery single guest felt itrolling the refreshed ride out to everyone — then it broke

The refreshed ride opened to everyone — then it broke.No matter how much you polish it, a new version can still break in front of everyone.

tested a hundred times, still not enoughin front of real guests, who knows what breaksnobody really knows until real guests show up

No matter how much you test, you cannot know until real guests show up.Rehearsal cannot catch everything. Only real guests give the real answer.

new version, tucked in the cornertesting quietly in a corner — just 1% of guestsif it is fine, add a little more — if not, pull it back right away

Quietly, in a corner, try it on just 1% of guests.If it's fine, add a little more. If not, pull it back right away.

Only a handful of guests firstquietly enough not to show
Keep watching the gaugesfor any sign of trouble
If it's fine, raise the share bit by bit1% → 10% → 50%
If not, pull it all back at oncerevert it
1%10%50%100%✓✓✓check the gauges at every stage before moving on

1% → 10% → 50% → 100% — check the gauges at every stage.Never jump to all at once — move to the next stage only after confirming it's fine.

there were only a few guests to begin with...1% doesn't even add up to one guestcanaries only work when 1% is still enough guests to notice

When there are only a few guests, 1% doesn't tell you much.If 1% doesn't even add up to one guest, the test run tells you nothing.

In one breath

Canary deployment = quietly trying a new version on just 1% of guests first, watching the gauges, and either scaling up slowly or pulling back immediately.

A gradual rollout strategy that exposes a new version to only a small slice of traffic first, reducing risk. If the metrics look fine, the share increases step by step; if they worsen, it's rolled back automatically.

When grown-ups say it

Canary deployment
A quiet test run in the corner. Only 1% of guests see the new version first.
Gradual rollout
Raising the share a little at a time. Never switching everything over all at once.
Traffic splitting
Sending guests down different paths. A slice goes to the new ride, the rest to the old one.
Metric comparison
Watching the new and old version's gauges side by side. Any difference shows up right away.
Automatic rollback
Reverting on its own when something's off. Fine even if a person notices late. → back to yesterday's version, fast
vs. blue-green
Switching everything, versus showing it to a few first. Blue-green switches the whole thing at once
Feature flag
A curtain drawn over the new ride. Paired with a canary, it picks exactly who sees the new ride.
Rollback
Back to yesterday's version, right away. → back to yesterday's version, fast
Next story
Running Twin Rides, One at a Time →
Want another term as a picture book? Request →