BEGIN:VCALENDAR VERSION:2.0 PRODID:-//HasGeek//NONSGML Funnel//EN DESCRIPTION:Hands-on workshop going from zero to a working LSM Tree in Go — Rootconf Topical Edition on Databases X-WR-CALDESC:Hands-on workshop going from zero to a working LSM Tree in Go — Rootconf Topical Edition on Databases NAME:Cascade — build a toy LSM Tree from scratch X-WR-CALNAME:Cascade — build a toy LSM Tree from scratch REFRESH-INTERVAL;VALUE=DURATION:PT12H SUMMARY:Cascade — build a toy LSM Tree from scratch TIMEZONE-ID:Asia/Kolkata X-PUBLISHED-TTL:PT12H X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT SUMMARY:Cascade — build a toy LSM Tree from scratch DTSTART:20260627T083000Z DTEND:20260627T123000Z DTSTAMP:20260610T121118Z UID:session/CxuwUU3exunJk4CYQtQS9b@hasgeek.com SEQUENCE:7 CREATED:20260608T061136Z DESCRIPTION:## 🎯 Target audience\n- Engineers with moderate systems pro gramming experience who want to understand how LSM Trees work internally — and why they are designed the way they are\n- No prior database intern als knowledge required\, but comfort with a systems programming language i s expected\n\n## 📘 Workshop overview\nThis workshop goes in depth on ho w a Log-Structured Merge Tree works — the storage engine architecture be hind RocksDB\, LevelDB\, Cassandra\, and Couchbase. By the end\, you will have written your own LSM Tree in Go that handles writes\, reads\, and del etions\; persists data to disk in SSTable format\; and runs a full tiered compaction pass across levels.\n\nThe central tension the workshop explore s: what are you paying for reads when writes are relatively cheap? Every d esign decision in an LSM Tree is a response to that question. You will fee l this tradeoff directly through a read IO counter that tracks how many IO s a read requires as data accumulates across levels.\n\n## ✅ Learning ou tcomes\nBy the end of the workshop\, participants will be able to understa nd:\n- The write-optimised tradeoff at the heart of LSM Trees\, and why it exists\n- The on-disk layout of an SSTable and how it is written and read \n- Why L0 is unsorted\, and what that costs on the read path\n- How tombs tones work and why deletes are non-trivial\n- How reads degrade as SSTable s accumulate\, and how compaction recovers them\n- The tiered compaction a lgorithm: what it costs in write amplification\, what it saves in implemen tation complexity\n- What real systems like LevelDB and RocksDB extend fro m this foundation\n\n## 🧩 Workshop stages\n**1. Setup**\nGet the skelet on repository running\, understand the testing harness\, and run the first milestone tests. Sets the feedback loop for the rest of the workshop.\n\n **2. Flush**\nPersist data to disk as an SSTable. Introduce the on-disk fo rmat and discuss encoding choices.\n\n**3. L0 Reads**\nImplement reads fro m L0. Introduce the read IO counter. See directly what a read costs when d ata is scattered across SSTables.\n\n**4. Checkpointing and Recovery**\nHa ndle restarts gracefully: checkpoint current state and recover it on start up.\n\n**5. Compaction in Isolation**\nUnderstand the compaction algorithm as a standalone problem — merging and sorting two SSTable streams\, han dling tombstones\, producing a new SSTable.\n\n**6. Tiered Compaction End- to-End**\nWire compaction into the engine. Implement a full tiered compact ion pass across L0\, L1\, and L2. Observe the reduction in read IOs as com paction runs.\n\n## 🛠 Tooling\nGo\, with third-party libraries for the in-memory map and disk encoding. A skeleton repository with sample data\, a load generation harness\, and per-milestone tests will be provided.\n\n# # 📚 Further reading\n- O'Neill et al. (1996) — [The Log-Structured Me rge-Tree](https://www.cs.umb.edu/~poneil/lsmtree.pdf) — the canonical pa per that introduced LSM Trees. Read this after the workshop to see how muc h of what you built was already in the original design.\n- [LevelDB source code](https://github.com/google/leveldb) — a direct\, readable implemen tation of the ideas from the workshop\, and the direct ancestor of RocksDB . Every component will look familiar.\n- [RocksDB Wiki: Compaction](https: //github.com/facebook/rocksdb/wiki/Compaction) — how a production system extends tiered and leveled compaction\, and the operational tradeoffs bet ween them.\n\n## About the instructor\n[**Anirudh Rowjee**](https://rowjee .com)\, Software Engineer — Storage @ Couchbase. \n\n## How to attend th is workshop\nThis workshop is open for **[Rootconf members](https://hasgee k.com/rootconf#memberships)** and for **[Rootconf Database Edition ticket buyers](https://hasgeek.com/rootconf/topical-edition-on-databases#tickets) **\n\n**This workshop is open to 30 participants (in-person) & hybrid acce ss for remote attendees. Seats for in-person participants will be availabl e on first-come-first-served basis.** 🎟️ \n\n## Contact informatio n ☎️\nFor inquiries about the workshop\, contact +91-7676332020 or wri te to [info@hasgeek.com](mailto:info@hasgeek.com). LAST-MODIFIED:20260608T102908Z LOCATION:Bangalore - https://hasgeek.com/rootconf/build-a-toy-lsm-tree-fro m-scratch-workshop/ ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/rootconf/build-a-toy-lsm-tree-from-scratch-worksho p/ BEGIN:VALARM ACTION:display DESCRIPTION:Cascade — build a toy LSM Tree from scratch in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT END:VCALENDAR