Layer 3 · Go deeper · 7 min
Go deeper
Size it, build it, then make the calls an interviewer will push you on.
Back of the envelope
Size it up
Tap through each step of the arithmetic.
How many pods does a 10-replica rolling update run, at most and least?
Can a 1% canary spot a release that doubles the error rate?
Build-up · 5 steps
Watch a canary fail its second gate
One canary of a new build, judged against a fresh baseline at every step and rolled back with nobody watching.
Step 1 of 5
Plan the steps, split by weight
The plan ramps 1% → 5% → 25% → 50% → 100%, pausing at each step. Weighted routing sends an exact share, and hashing the user ID keeps each user on one version.
4 interview moments
You’re the candidate
The interviewer pushes on the hard parts. Pick an approach and see how it holds up.
Deep dive 1 of 4
Interviewer
“Checkout is busy, and a bad build is expensive. How do you put the new version in front of users?”
Your move. Pick an approach:
Deep dive 2 of 4
Interviewer
“Your quiet service gets 50 requests a second. The plan starts with a 1% canary for 10 minutes. Is that enough?”
Your move. Pick an approach:
Deep dive 3 of 4
Interviewer
“v2 starts compressing everything it stores. How do you ship it so rollback stays safe?”
Your move. Pick an approach:
Deep dive 4 of 4
Interviewer
“A teammate wants to switch a new feature’s flag on for every server at once. “It’s just config.” What do you say?”
Your move. Pick an approach: