Patterns of Distributed Systems

Unmesh Joshi is a Principal Consultant at Thoughtworks with 22 years of industry experience. He is a software architecture enthusiast, who believes that understanding principles of distributed systems is as essential today as understanding web architecture or object-oriented programming was in the last decade. For the last two years he has been publishing patterns of distributed systems on martinfowler.com. He has also conducted various training sessions around this topic. Twitter: @unmeshjoshi

• Foreword xviiPreface xixAcknowledgments xxiiiAbout the Author xxvPart I: Narratives 1Chapter 1: The Promise and Perils of Distributed Systems 3The Limits of a Single Server 3Separate Business Logic and Data Layer 5Partitioning Data 6A Look at Failures 7Replication: Masking Failures 9Defining the Term "Distributed Systems" 10The Patterns Approach 10Chapter 2: Overview of the Patterns 13Keeping Data Resilient on a Single Server 14Competing Updates 15Dealing with the Leader Failing 17Multiple Failures Need a Generation Clock 21Log Entries Cannot Be Committed until They Are Accepted by a Majority Quorum 26Followers Commit Based on a High-Water Mark 29Leaders Use a Series of Queues to Remain Responsive to Many Clients 34Followers Can Handle Read Requests to Reduce Load on the Leader 40A Large Amount of Data Can Be Partitioned over Multiple Nodes 42Partitions Can Be Replicated for Resilience 45A Minimum of Two Phases Are Needed to Maintain Consistency across Partitions 46In Distributed Systems, Ordering Cannot Depend on System Timestamps 49A Consistent Core Can Manage the Membership of a Data Cluster 58Gossip Dissemination for Decentralized Cluster Management 62Part II: Patterns of Data Replication 69Chapter 3: Write-Ahead Log 71Problem 71Solution 71Examples 76Chapter 4: Segmented Log 77Problem 77Solution 77Examples 79Chapter 5: Low-Water Mark 81Problem 81Solution 81Examples 83Chapter 6: Leader and Followers 85Problem 85Solution 85Examples 92Chapter 7: HeartBeat 93Problem 93Solution 93Examples 98Chapter 8: Majority Quorum 99Problem 99Solution 100Examples 102Chapter 9: Generation Clock 103Problem 103Solution 104Examples 107Chapter 10: High-Water Mark 109Problem 109Solution 109Examples 115Chapter 11: Paxos 117Problem 117Solution 117Examples 132Chapter 12: Replicated Log 133Problem 133Solution 133Examples 158Chapter 13: Singular Update Queue 159Problem 159Solution 159Examples 166Chapter 14: Request Waiting List 167Problem 167Solution 167Examples 173Chapter 15: Idempotent Receiver 175Problem 175Solution 175Examples 181Chapter 16: Follower Reads 183Problem 183Solution 183Examples 191Chapter 17: Versioned Value 193Problem 193Solution 193Examples 201Chapter 18: Version Vector 203Problem 203Solution 203Examples 216Part III: Patterns of Data Partitioning 217Chapter 19: Fixed Partitions 219Problem 219Solution 220Examples 241Chapter 20: Key-Range Partitions 243Problem 243Solution 244Examples 255Chapter 21: Two-Phase Commit 257Problem 257Solution 257Examples 297Part IV: Patterns of Distributed Time 299Chapter 22: Lamport Clock 301Problem 301Solution 301Examples 307Chapter 23: Hybrid Clock 309Problem 309Solution 309Examples 316Chapter 24: Clock-Bound Wait 317Problem 317Solution 318Examples 332Part V: Patterns of Cluster Management 335Chapter 25: Consistent Core 337Problem 337Solution 337Examples 342Chapter 26: Lease 345Problem 345Solution 345Examples 354Chapter 27: State Watch 355Problem 355Solution 355Examples 362Chapter 28: Gossip Dissemination 363Problem 363Solution 363Examples 373Chapter 29: Emergent Leader 375Problem 375Solution 375Examples 392Part VI: Patterns of Communication between Nodes 393Chapter 30: Single-Socket Channel 395Problem 395Solution 395Examples 397Chapter 31: Request Batch 399Problem 399Solution 399Examples 404Chapter 32: Request Pipeline 405Problem 405Solution 405Examples 408References 409Index 413