Architectural Case Studies & System Postmortems
In-depth breakdowns of production architectures, consensus protocol bottlenecks, extreme database migrations, and high-throughput infrastructure running at scale.
Designing a Distributed Rate Limiter with Redis & Lua at 100,000 QPS
A comprehensive teardown of coordinating sub-millisecond rate limits across multi-region edge gateways using sliding window counter algorithms in Redis cluster.
Coordinating high-frequency API rate limiting across 12 edge regions without lock contention or redis thread blocking.
Atomic Lua scripts executing sliding-log counters with local sliding window approximations.
100k QPS
Throughput
p99 < 1.8ms
Latency
99.999%
Availability
Zero-Downtime PostgreSQL Schema Migrations on a 5TB Database
Safe multi-step column additions, concurrent index creation, avoiding AccessExclusiveLocks, and maintaining backward-compatible ORM contracts during live traffic.
Altering large tables with hundreds of millions of rows caused query timeouts due to exclusive lock queues.
Expand-contract migration pipeline using shadow columns, trigger synchronization, and concurrent indexing.
5 TB DB
Data Volume
0.00s
Downtime
< 50ms
Lock Wait
Kafka Partitioning & Consumer Rebalancing for Zero Data Loss
Eliminating message duplication and out-of-order execution in financial transaction pipelines using custom partitioners and cooperative sticky assignors.
Eager partition rebalancing caused consumer stop-the-world pauses and duplicate event processing.
CooperativeStickyAssignor with transactional producer IDs and exactly-once processing semantics.
0.00%
Message Loss
45k msgs/s
Event Rate
< 200ms
Rebalance Lag