Layer 3 · Go deeper · 9 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.
What share of that budget does each incident spend?
Which burn rates should page, and which should open a ticket?
Build-up · 5 steps
Spend a 30-day budget
A 99.9% SLO, 50 million requests a month. A small error rate drips steadily; three incidents take big bites, and the policy reacts.
Step 1 of 5
50,000 failures to spend
99.9% of 50 million requests leaves 50,000 failures. The dashed line is a burn rate of 1: on pace to reach zero exactly on day 30.
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
“You need an availability SLI for order tracking. Web server logs exist today. Where do you measure it first?”
Your move. Pick an approach:
Deep dive 2 of 4
Interviewer
“Restaurant search runs on many servers, and product wants it fast. How do you write the latency SLO?”
Your move. Pick an approach:
Deep dive 3 of 4
Interviewer
“A service has a 99.9% SLO over 30 days. On-call wants a page only when it matters. What do you alert on?”
Your move. Pick an approach:
Deep dive 4 of 4
Interviewer
“The pay API needs its first SLO. What do you propose?”
Say the last four weeks show 99.968% success, a p90 of 412 ms and a p99 of 1,180 ms.
Your move. Pick an approach:
Build-up · 5 steps
Page fast, then stop fast
A bad deploy fails 10% of requests for 15 minutes against a 99.9% SLO. Watch which windows cross 14.4×, assuming steady traffic.
Step 1 of 5
A bad deploy
A deploy fails 10% of requests, a burn rate of 100 against a 99.9% SLO, until it is rolled back at minute 15.