System Design: 15 Lessons From Real Outages
Requests, APIs, databases, queues, scaling, resilience, caching, cost, plus the interview playbook.
Forensic teardowns of production systems: distributed systems, AI infrastructure, telemetry pipelines and cloud bills, down to the cost per request. Every claim reconstructed from public disclosures, every model runnable in your browser.
By Mukul Kumar Mishra with Rook research desk · About the publication
Golang plus AWS at scale? Architecture reviews plus SRE audits → Work with Mukul
A practical teardown of WebSocket collaboration, Rust isolation, conflict resolution plus the hidden price of an always-live canvas.
Proxy bypass first plus containment second. How tail-matched exceptions fail plus what identity-bound egress costs, modeled from the wiki swarm.
Channel partitioning, quorum read fan-in, Rust coalescing plus the engine swap from 177 nodes to 72.
Eager versus cooperative protocols, static membership, fetch storms plus the 6x retry amplifier.
How Figma multiplayer works: WebSocket fan-out, server-authoritative resolution, Rust rewrite plus cost at scale.
How Notion scales: 480 Postgres shards, CDC lake, vector tiering plus the bill behind a blank page.
A forensic teardown of RAG retrieval, KV cache flaws, prefill tax, 780M monthly queries plus the cost difference at 1M events per second.
A deep teardown of prompt-to-production software, sandbox execution, agent retries, deployment risk plus the cost difference at 1M events per second.
A forensic look at concurrency, cold starts, compute-resolver, AI Gateway waiting time plus the cost difference at 1M RPS.
A practical teardown of WebSocket collaboration, Rust isolation, conflict resolution plus the hidden price of an always-live canvas.
A forensic teardown of Notion's block database, 480 logical shards, CDC pipeline, vector search plus modeled cloud bill.
A control-plane redesign focused on regional isolation, failure containment, Raft plus zero-blast-radius operations.
Requests, APIs, databases, queues, scaling, resilience, caching, cost, plus the interview playbook.
Sandboxing, egress, retries, token bills plus eval gates. From the agent bench, with live diagrams.
Control planes, autoscaling, blast radius plus inference scheduling. From the on-call, with live diagrams.