Table of Contents
Preface
How to Contact Us
I. Storage Engines
1. Introduction and Overview
DBMS Architecture
Memory- Versus Disk-Based DBMS
Durability in Memory-Based Stores
Column- Versus Row-Oriented DBMS
Row-Oriented Data Layout
Column-Oriented Data Layout
Distinctions and Optimizations
Wide Column Stores
Data Files and Index Files
Data Files
Index Files
Primary Index as an Indirection
Buffering, Immutability, and Ordering
Summary
2. B-Tree Basics
Binary Search Trees
Tree Balancing
Trees for Disk-Based Storage
Disk-Based Structures
Hard Disk Drives
Solid State Drives
On-Disk Structures
Ubiquitous B-Trees
B-Tree Hierarchy
Separator Keys
B-Tree Lookup Complexity
B-Tree Lookup Algorithm
Counting Keys
B-Tree Node Splits
B-Tree Node Merges
Summary
3. File Formats
Motivation
Binary Encoding
Primitive Types
Strings and Variable-Size Data
Bit-Packed Data: Booleans, Enums, and Flags
General Principles
Page Structure
Slotted Pages
Cell Layout
Combining Cells into Slotted Pages
Managing Variable-Size Data
Versioning
Checksumming
Summary
4. Implementing B-Trees
Page Header
Magic Numbers
Sibling Links
Rightmost Pointers
Node High Keys
Overflow Pages
Binary Search
Binary Search with Indirection Pointers
Propagating Splits and Merges
Breadcrumbs
Rebalancing
Right-Only Appends
Bulk Loading
Compression
Vacuum and Maintenance
Fragmentation Caused by Updates and Deletes
Page Defragmentation
Summary
5. Transaction Processing and Recovery
Buffer Management
Caching Semantics
Cache Eviction
Locking Pages in Cache
Page Replacement
Recovery
Log Semantics
Operation Versus Data Log
Steal and Force Policies
ARIES
Concurrency Control
Serializability
Transaction Isolation
Read and Write Anomalies
Isolation Levels
Optimistic Concurrency Control
Multiversion Concurrency Control
Pessimistic Concurrency Control
Lock-Based Concurrency Control
Summary
6. B-Tree Variants
Copy-on-Write
Implementing Copy-on-Write: LMDB
Abstracting Node Updates
Lazy B-Trees
WiredTiger
Lazy-Adaptive Tree
FD-Trees
Fractional Cascading
Logarithmic Runs
Bw-Trees
Update Chains
Taming Concurrency with Compare-and-Swap
Structural Modification Operations
Consolidation and Garbage Collection
Cache-Oblivious B-Trees
van Emde Boas Layout
Summary
7. Log-Structured Storage
LSM Trees
LSM Tree Structure
Updates and Deletes
LSM Tree Lookups
Merge-Iteration
Reconciliation
Maintenance in LSM Trees
Read, Write, and Space Amplification
RUM Conjecture
Implementation Details
Sorted String Tables
Bloom Filters
Skiplist
Disk Access
Compression
Unordered LSM Storage
Bitcask
WiscKey
Concurrency in LSM Trees
Log Stacking
Flash Translation Layer
Filesystem Logging
LLAMA and Mindful Stacking
Open-Channel SSDs
Summary
Part I Conclusion
II. Distributed Systems
8. Introduction and Overview
Concurrent Execution
Shared State in a Distributed System
Fallacies of Distributed Computing
Processing
Clocks and Time
State Consistency
Local and Remote Execution
Need to Handle Failures
Network Partitions and Partial Failures
Cascading Failures
Distributed Systems Abstractions
Links
Two Generals’ Problem
FLP Impossibility
System Synchrony
Failure Models
Crash Faults
Omission Faults
Arbitrary Faults
Handling Failures
Summary
9. Failure Detection
Heartbeats and Pings
Timeout-Free Failure Detector
Outsourced Heartbeats
Phi-Accural Failure Detector
Gossip and Failure Detection
Reversing Failure Detection Problem Statement
Summary
10. Leader Election
Bully Algorithm
Next-In-Line Failover
Candidate/Ordinary Optimization
Invitation Algorithm
Ring Algorithm
Summary
11. Replication and Consistency
Achieving Availability
Infamous CAP
Use CAP Carefully
Harvest and Yield
Shared Memory
Ordering
Consistency Models
Strict Consistency
Linearizability
Sequential Consistency
Causal Consistency
Session Models
Eventual Consistency
Tunable Consistency
Witness Replicas
Strong Eventual Consistency and CRDTs
Summary
12. Anti-Entropy and Dissemination
Read Repair
Digest Reads
Hinted Handoff
Merkle Trees
Bitmap Version Vectors
Gossip Dissemination
Gossip Mechanics
Overlay Networks
Hybrid Gossip
Partial Views
Summary
13. Distributed Transactions
Making Operations Appear Atomic
Two-Phase Commit
Cohort Failures in 2PC
Coordinator Failures in 2PC
Three-Phase Commit
Coordinator Failures in 3PC
Distributed Transactions with Calvin
Distributed Transactions with Spanner
Database Partitioning
Consistent Hashing
Distributed Transactions with Percolator
Coordination Avoidance
Summary
14. Consensus
Broadcast
Atomic Broadcast
Virtual Synchrony
Zookeeper Atomic Broadcast (ZAB)
Paxos
Paxos Algorithm
Quorums in Paxos
Failure Scenarios
Multi-Paxos
Fast Paxos
Egalitarian Paxos
Flexible Paxos
Generalized Solution to Consensus
Raft
Leader Role in Raft
Failure Scenarios
Byzantine Consensus
PBFT Algorithm
Recovery and Checkpointing
Summary
Part II Conclusion
A. Bibliography
Index