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.
How long does restoring a 10 TB database take?
What does one write from Frankfurt to a Mumbai primary cost?
Build-up · 5 steps
Follow order 9182 through one server
A food-delivery app asks for the status of order 9182. Watch which component handles each hop, and which limit lives there.
Step 1 of 5
Connect
The app connects, directly or through an app server. The communications manager speaks the wire protocol, authenticates and will buffer the result rows. Too many connections strain it first.
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
“Your site, under 100K hits a day, runs on SQLite. A second app server is coming. Where does the data live now?”
Your move. Pick an approach:
Deep dive 2 of 4
Interviewer
“A write-heavy SaaS with a natural tenant key has outgrown the biggest primary you can buy. What do the machines share?”
Your move. Pick an approach:
Deep dive 3 of 4
Interviewer
“A read-heavy service on AWS needs fast failover and about 60,000 reads a second. What do you build?”
Your move. Pick an approach:
Deep dive 4 of 4
Interviewer
“Users in India and Europe write shared data and need strongly consistent transactions across both. Your primary is in Mumbai. What do you do?”
Your move. Pick an approach:
Build-up · 5 steps
Six copies, four to write, three to read
One Aurora writer and a 10 GB segment of its volume, stored six times across three availability zones, two per zone.
Step 1 of 5
Only redo crosses the network
The writer sends redo log records to all six copies, never pages, not for checkpoints and not for cache eviction. Storage nodes apply the records and build pages in the background.