Interview with – Vadim Tsesko
summary
Built an internal S3-compatible object store that serves exabytes of blobs with hot/cold tiers, erasure coding and replication across three datacenters, and handles peaks around 50 Gbps; metadata lives in Cassandra-backed metadata with bespoke transactional behavior added; deployment uses partitioned feature flags (256 partitions) and gradual rollouts, and testing runs against a local S3 stack. An accidental global debug flag once brought Cassandra nodes down, so operational safety and rollout tooling earn heavy attention; open sourcing is not planned now.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 3:31Teamwork and design patterns
- 10:38Cross-platform issues and protocol
- 18:52Building S3-compatible storage from scratch
- 25:15Storage system challenges and optimizations
- 31:17Open resources and education
- 39:03Project future and roadmap
- 44:01Motivation and closing
related talks
27mRebuilding a Faster OSD with Future
34mCrimson Project Update
06mGlobal Scale Multi-Cloud, Multi-CDN Load Balancing with C++
59mMaking Testing C++ Binaries Practical @ Facebook Scale: A CI Story
30m“Esoteric Data Structures and Where to Find Them”
30mA New Open Source Virtual File System – running on Windows in User mode!
01h 04mBetter Code: Data Structures
56mLightning Updates