Distributed blob storage built on SQLite
daily.dev

Distributed blob storage built on SQLite

Archived — this story has rotated out of today’s deck. It is kept here in full.

The gist

Indian brokerage Zerodha released Blobasaur, an open-source Rust blob storage server built on SQLite.
It replaces an unmaintained single-node store and now serves millions of Kite users.

Background

Zerodha is an Indian brokerage whose engineering team runs Kite, a trading platform with about 20 million customers. Its user preference blobs had moved from CouchDB to SSDB, a C++ disk-backed store speaking the Redis protocol that ran reliably for years but became unmaintained and only worked on a single node. Blobasaur, written in Rust on top of SQLite and speaking the Redis and Redis Cluster protocols, was built to replace SSDB and scale past one machine as features like Kite's Terminal mode raised load. Sarat Chandra, a Zerodha software engineer, described it in a blog post, and Kailash Nadh announced it on LinkedIn.

How it unfolded

  1. Nov 19, 2026Zerodha Tech Blog publishes "Distributed, high-performance blob storage on SQLite" by engineer Sarat Chandra, describing Blobasaur.
  2. Nov 19, 2026Kailash Nadh says on LinkedIn that Blobasaur was built, tested and deployed over a year after no simple-to-run database tailored for blob storage was found.
  3. Nov 19, 2026Reported benchmarks: 16 shards raised HSET write throughput 3.3x over one shard, from 7,740 to 25,324 ops/s, with median write latency falling from 8.2 ms to 2.1 ms.
  4. Nov 19, 2026Zstd compression cuts production payloads roughly 11x on average; on a three-node production cluster p99 latency stays around 2 ms.
  5. Nov 19, 2026Coverage and discussion follow on daily.dev, LinkedIn, TLDR, hckrnws and Lobsters.

Key terms

blob
A data storage approach that manages data as objects or blobs rather than as files or blocks.
SQLite
A free, open-source relational database engine written in C that is embedded as a library inside applications rather than run as a standalone program.

Source: Wikipedia

Who’s saying what

Official
Zerodha's engineering team says SQLite was chosen because it is up to 35% faster than raw filesystem I/O for small blobs and brings built-in ACID transactions and crash recovery without extra operational overhead.
Official
Kailash Nadh says Blobasaur handles tens of thousands of ops/sec on commodity hardware with negligible resource use and single-digit millisecond latency, calling the design "least-wrong tradeoffs".
Caution
A LinkedIn commenter disputes the blog's characterization of RocksDB, saying its BlobDB does key-value separation and that the blog does not clearly measure multi-MB data.
Public
daily.dev commenters were largely skeptical of using SQLite for this workload, pointing to RocksDB, rqlite, ClickHouse, Ceph and plain files as alternatives.

Still unverified

daily.a LinkedIn commenter claims the blog's premise about RocksDB is false and that multi-MB writes are not clearly measured.

Sources

See today’s stories in the app Gistline — free on the App Store