Design brief
Design delivery for public product images and translated help pages at 20,000 peak requests per second with a 200 KB average body. Direct origin traffic is roughly 4 GB per second. Accept up to 60 seconds of ordinary staleness and an additional 30 seconds during refresh for selected public pages. Personalized account responses must remain private. Define latency goals and a publication-freshness contract.
Draw and explain
In Excalidraw, show reader routing, edge boundaries, origin ownership, durable content, and the management path for publication or purge. Label the full cache identity, response variants, validators, freshness deadlines, and access checks. Trace a fresh hit, an expired popular entry, and a publisher update. Show how origin request and byte budgets change with the expected hit ratio.
Stress the design
A thousand readers arrive as one popular entry expires while the origin slows. Explain refresh ownership, waiting or stale responses, and the hard limit on staleness. Then let a purge race with an older in-flight fill. Draw the ordering evidence that prevents old content from becoming current again.
Review your design
Check language and query variants, authenticated traffic, and bounded cache-key cardinality. Compare versioned URLs with distributed purges for different content. Explain the cost of more variants, regional replication, and stale serving. Choose signals for byte hit ratio, origin amplification, purge propagation, and regional latency.
Capture assumptions, sizing, and the tradeoffs behind your diagram.
Review is your own assessment of the design.