Arpit Mishra
Fullstack AI Developer
Socials
Choosing, sharding, replicating and querying data at scale
When a single transaction has to span multiple services and databases, you can't just wrap it in BEGIN/COMMIT — here's how 2PC and sagas each try to fake atomicity, and why one of them lost.
Consistency isn't a checkbox — it's a spectrum of promises about what a reader is allowed to see, and each point on it costs something different.
Why an index turns a table scan into a lookup, what it costs on every write, and how B-Trees, LSM trees, and hash indexes each trade that off differently.
The SQL-vs-NoSQL debate isn't really one question — it's schema, joins, scaling, and transactions bundled together, and each deserves its own answer.
Why we copy data across machines, the three ways to do it, and the theorem that explains what you give up the moment the network fails.
Partitioning and sharding get used interchangeably, but only one of them buys you more write capacity — here's the distinction, and the checklist for when sharding is actually the right call.
No master, no single point of failure — how Cassandra's ring, LSM-tree write path, and tunable consistency add up to a database built for write-heavy, always-on workloads.
Forget the feature checklist — the right database is the one whose access patterns match yours, not the one with the most familiar logo.
