Bookmarks
A curated collection of useful reading resources I've found interesting.
51 bookmarks across 5 categories
Database Internals
Hot partitions, compaction, JVM pauses, request coalescing, and a zero-downtime Cassandra-to-Scylla migration.
How connection limits and replication constraints led Figma to a horizontally scalable Postgres architecture.
A candid account of the storage and operational constraints that force a Postgres sharding migration.
The mechanics and safeguards for adding Postgres capacity with a live, zero-downtime reshard.
Partitioning trade-offs where foreign keys, joins, and application ownership meet real database limits.
A practical migration of a high-traffic relational workload onto Vitess sharding.
How a large multi-tenant SaaS moved its MySQL scaling boundary with Vitess.
Mechanism-level diagnostics for dead tuples, bloat, vacuum progress, and runaway maintenance debt.
A PostgreSQL committer's guide to the estimation errors that create unexpected bad plans.
How heavy partitioning can defeat fast-path locking and create a lock-manager scalability cliff.
An obscure but important write-path bottleneck caused by shared row locks and foreign keys.
How an unconsumed replication slot retains WAL until disk becomes the outage mechanism.
The core row-locking primitive behind safe multi-worker database queues and schedulers.
Storage-engine trade-offs in read, write, space, and compaction amplification.
Distributed Systems
The classic explanation of why cron breaks at company scale and what a distributed scheduler must add.
How a durable workflow engine protects fair task processing when tenants have very different loads.
A production workflow-orchestration design based on task queues and JSON-defined workflows.
The architecture of durable execution: history, matching, persistence, and independently scaled task queues.
The canonical ephemeral-sequential-node algorithm for leader election and failure detection.
A step-by-step reconciliation loop for determining due, running, and expired scheduled work.
A real priority-queue architecture designed for high throughput without sacrificing latency.
Disaster-readiness techniques for a distributed priority queue that must survive regional failures.
Performance & Debugging
Systematic performance analysis methodology examining system resources for Utilization, Saturation, and Errors to quickly identify bottlenecks.
The definitive guide to finding hot code paths from CPU, memory, off-CPU, and differential stack profiles.
The longer-form performance-investigation method behind USE, with practical examples.
Why busy CPU percentage can hide the real limiter, and how to reason about IPC instead.
A production disk-latency investigation that applies resource analysis from symptom to storage tier.
The foundational deep dive into caches, NUMA, TLBs, and the memory hierarchy.
Kernel documentation on cache-line contention that silently destroys multicore throughput.
A tools-first kernel debugging investigation using tracing to separate application error from a Linux race.
Twelve cache failures distilled into recurring patterns: retries, cascading failures, and hardware surprises.
JVM heap and garbage-collection tuning tied directly to a service throughput target.
How JVM heap and GC choices shape latency SLOs in production services.
Object-store and disk-level garbage-collection mechanics at platform scale.
A concrete capacity and latency story for running an unusually large real-time workload.
Storage & Caching
A decade of storage, partitioning, and operational lessons from a managed key-value store.
How Meta handles invalidation races, thundering herds, and cache/database consistency at scale.
Production chain replication: its failure model, consistency guarantees, and storage-service design.
RocksDB-based storage-engine choices wrapped in a distributed key-value architecture.
How cache behavior and hit ratios inform an online-storage architecture at extreme read volume.
Compaction, JVM, and disk-I/O work that turns Cassandra operations into a manageable fleet concern.
A search-service bootstrap redesign that reduces Redis pressure with local RocksDB, LSM tuning, offset recovery, and native-memory debugging.
The design of Redis lazy freeing: avoiding delete-induced latency spikes while balancing memory pressure, cache misses, and threading.
A first-principles explanation of write, fsync, snapshots, AOF, and the durability guarantees a database can actually make.
A high-availability data-migration playbook built around document databases and operational safety.
A consistent metadata cache designed to make cache invalidation safe rather than merely fast.
Why PayPal built a key-value store and the replication and consistency decisions behind it.
System Design
In-depth technical paper on Amazon DynamoDB, a fully managed NoSQL database service.
In-depth exploration of the Kafka communication protocol, including its architecture and design principles.
AWS's mathematical approach to isolating tenants and reducing correlated failure blast radius.
A rigorous guide to retry storms, overload feedback loops, and the limits that stop them.