BEGIN:VCALENDAR VERSION:2.0 PRODID:-//HasGeek//NONSGML Funnel//EN DESCRIPTION:Design MVCC databases with real-world consistency guarantees X-WR-CALDESC:Design MVCC databases with real-world consistency guarantees NAME:Building a distributed transactional database X-WR-CALNAME:Building a distributed transactional database REFRESH-INTERVAL;VALUE=DURATION:PT12H SUMMARY:Building a distributed transactional database TIMEZONE-ID:Asia/Kolkata X-PUBLISHED-TTL:PT12H X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT SUMMARY:Building a distributed transactional database DTSTART:20260328T043000Z DTEND:20260328T110000Z DTSTAMP:20260415T000403Z UID:session/6HKPTdhunqeoqQhr8xLJ1q@hasgeek.com SEQUENCE:17 CREATED:20260204T071512Z DESCRIPTION:# 🧱 Workshop: Building a Distributed Transactional Database \n\n## 🎯 Target Audience\n\n* Senior engineers with a background in dis tributed systems\, transactions\, and concurrency control\n* Enthusiasts w ho want to understand how transactions work in distributed databases\n\n-- -\n\n## πŸ“˜ Abstract\n\nThis workshop explores the design space of global ly consistent transactional databases by incrementally building a distribu ted **Multi-Version Concurrency Control (MVCC)** key-value store using **H ybrid Logical Clocks (HLC)**.\n\nParticipants will implement **Snapshot Is olation (SI)** and examine how **strongly consistent reads** can be achiev ed under bounded clock skew without hardware-assisted time like TrueTime\, using uncertainty detection and restart techniques.\n\nThe workshop empha sizes:\n\n* Correctness invariants\n* Timestamp ordering\n* Trade-offs bet ween commit-wait and retry-based designs\n\nIt provides a unified conceptu al framework for understanding systems such as **Google Spanner\, Cockroac hDB\, and YugabyteDB**. \n\n---\n\n## βœ… Learning outcomes\n\nBy the end of the workshop\, participants will be able to:\n\n* Reason about time\, o rdering\, and causality in distributed databases\n* Implement Hybrid Logic al Clocks\n* Build an MVCC storage engine on top of a RocksDB-like store\n * Implement Snapshot Isolation\n* Achieve linearizable reads under bounded clock skew\n* Understand the limits of clock-based approaches\n\n---\n\n# # πŸ“‹ Prerequisites\n\n* Java 21+\n* Basic understanding of distributed s ystems\n\n---\n\n## πŸ”— Resources\n\n* **Repository:** [https://github.co m/unmeshjoshi/hybridclocktxn](https://github.com/unmeshjoshi/hybridclocktx n)\n* **Framework:** Tickloom – [https://github.com/unmeshjoshi/tickloom ](https://github.com/unmeshjoshi/tickloom)\n *(Used for determinism and s imulation testing)*\n\n---\n\n# 🧩 Workshop Modules\n\n---\n\n## Module 1: Time and Order in Distributed Systems (60 minutes)\n\n### 🎯 Objectiv e\n\nUnderstand why physical clocks fail and implement Hybrid Logical Cloc ks.\n\n### πŸ“š Concepts\n\n* Physical clock drift and NTP limitations\n* Logical clocks (Lamport) and causality\n* Hybrid Logical Clocks (HLC): com bining physical time with logical ticks\n\n### πŸ’» Coding Tasks\n\n#### 1 . Implement `HybridTimestamp`\n\nValue object holding:\n\n* `wallClockTime `\n* `logicalTicks`\n* Comparison logic\n\n#### 2. Implement `HybridClock` \n\n* `now()`\n* Handling physical time updates\n* Handling same-milliseco nd logical ticks\n* `tick(remoteTime)` (Lamport rule)\n* MAX_OFFSET sanity checks\n\n### πŸ“ Artifacts\n\n* `src/main/java/org/example/txn/HybridTi mestamp.java`\n* `src/main/java/org/example/txn/HybridClock.java`\n\n---\n \n## Module 2: The Storage Layer with RocksDB (60 minutes)\n\n### 🎯 Obj ective\n\nBuild a versioned key-value store using RocksDB.\n\n### πŸ“š Con cepts\n\n* LSM Trees and RocksDB basics\n* MVCC mapping\n* Key design: `Us erKey + Timestamp (descending) β†’ Value`\n* Why descending timestamps ena ble efficient latest-version seeks\n* Iterators\, scan\, reverseScan\n\n## # πŸ’» Coding Tasks\n\n#### 1. Implement `MVCCStore`\n\n* `put(key\, times tamp\, value)`\n β†’ `rocksDB.put(serialize(key\, timestamp)\, value)`\n\ n* `get(key\, readTimestamp)`\n β†’ `rocksDB.seek(serialize(key\, readTim estamp))`\n\n**Challenge:** efficient memcomparable key encoding.\n\n#### 2. Refactor\n\nReplace `ConcurrentSkipListMap` with `RocksDBManager`.\n\n# ## πŸ“ Artifacts\n\n* `src/main/java/org/example/txn/MVCCStore.java`\n* ` src/main/java/org/example/txn/RocksDBStorage.java`\n\n---\n\n## Module 3: Transactions and Snapshot Isolation (60 minutes)\n\n### 🎯 Objective\n\n Implement transaction lifecycle and Snapshot Isolation.\n\n### πŸ“š Concep ts\n\n* Snapshot reads at `T_start`\n* Write intents and buffering\n* Tran saction states: Running\, Committed\, Aborted\n* Atomic commit and visibil ity\n\n### πŸ’» Coding Tasks\n\n#### Read Path\n\n* Read at `T_read`\n* Re solve intents:\n\n * If intent is mine β†’ read\n * If intent is other:\ n\n * Committed < `T_read` β†’ visible\n * Committed > `T_read` β†’ ignore\n * Aborted / Pending β†’ ignore\n\n#### Write Path\n\n* Write i ntent with TxnId\n* Detect write–write conflicts\n* Check committed vers ions newer than `T_start`\n\n#### Commit Path\n\n* `commit(txnId)`\n* Mark COMMITTED with `T_commit`\n* Apply intents asynchronously\n\n### πŸ“ Art ifacts\n\n* `TxnClient.java`\n* `TransactionRecord.java`\n\n---\n\n## Modu le 4: Distributed Consistency β€” Clock-Bound Wait (60 minutes)\n\n### 🎯 Objective\n\nAchieve external consistency using HLC.\n\n### πŸ“š Conc epts\n\n* Clock skew & uncertainty window\n\n```\nT_now ∈ [T_now - Ξ΅\, T_now + Ξ΅]\n```\n\n* Why causal transactions may appear in the future\n* Commit-wait vs restart strategies\n\n### πŸ’» Coding Tasks\n\n1. Simulate uncertainty (`MAX_CLOCK_SKEW`)\n2. Detect commits in uncertainty window\n\ n```\nT_commit ∈ [T_read - Ξ΅\, T_read]\n```\n\n3. Trigger read restart with `T_restart = T_commit + Ξ΅`\n4. Client retry logic\n\n### πŸ“ Artifa cts\n\n* `StorageReplica.java` (handleRead updates)\n* `TxnClient.java` (r etry logic)\n\n---\n\n## Module 5: Putting It All Together (30 minutes)\n\ n### 🎯 Objective\n\nRun end-to-end simulations.\n\n### β–Ά Tasks\n\n* R un `StorageReplicaCommitTest` (Snapshot Isolation)\n* Run `ClockBoundWaitT est` (External Consistency)\n\n### About the instructor\n**Unmesh Joshi** is a Distinguished Engineer at Thoughtworks. He is a software architecture enthusiast\, who believes that understanding principles of distributed sy stems is as essential today as understanding web architecture or object-or iented programming was in the last decade. For the last two years he has b een publishing patterns of distributed systems on martinfowler.com. \n**In 2023\, he authored the [book Patterns of Distributed Systems](https://www .amazon.com/Patterns-Distributed-Systems-Addison-wesley-Signature/dp/01382 21987) published by Addison Wesley Professional.** This book is an essenti al catalog of patterns aimed at enhancing comprehension\, communication an d education on distributed system design\nHe has also conducted various tr aining sessions around this topic. Twitter: [@unmeshjoshi](https://x.com/u nmeshjoshi)\n\n### How to attend this workshop\nThis workshop is open for participation to [Rootconf annual members](https://hasgeek.com/rootconf#me mberships).\nThis workshop is open to 30 participants only. Seats will be available on first-come-first-serve basis. 🎟️\n\n### Contact informat ion ☎️\nFor inquiries about the workshop\, contact +91-7676332020 or w rite to info@hasgeek.com LAST-MODIFIED:20260413T053744Z LOCATION:Pune - https://hasgeek.com/rootconf/building-a-distributed-transa ctional-database/ ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/rootconf/building-a-distributed-transactional-data base/ BEGIN:VALARM ACTION:display DESCRIPTION:Building a distributed transactional database in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT END:VCALENDAR