Bigtable is composed of five main modules working together to provide distributed, large-scale structured storage:

• Master Server — Coordinates tablet assignments and handles metadata.
• Tablet Servers — Store and serve data; each handles a subset of tablets.
• GFS (Google File System) — Stores commit logs and SSTable files.
• Chubby — Provides distributed locking and master election.
• Client Library — Interfaces directly with tablet servers for read/write operations.

The Master controls the overall system but stays out of the data path to avoid bottlenecks. It assigns tablets, monitors server health, and handles tablet splits and recovery.

• Manages metadata and tablet distribution.
• Detects and reassigns tablets from failed servers.
• Coordinates load balancing across the cluster.

Tablet Servers manage tablets, which are continuous ranges of rows from a table. They serve client read/write requests directly and handle compactions and splits automatically.

• Each tablet stored as commit log + SSTables + memtable.
• Handles user queries and updates efficiently.
• Splits tablets when they grow too large.

The Google File System provides the underlying file storage for Bigtable. It ensures durability and fault tolerance for all persistent data.

• Stores immutable SSTable files and commit logs.
• Automatically replicates data across multiple machines.
• Facilitates fast recovery from hardware failures.

Chubby is a distributed lock service that Bigtable uses for coordination, master election, and metadata storage.

• Ensures only one active master at a time.
• Stores critical configuration and tablet location metadata.
• Provides lightweight synchronization for servers.

Clients interact directly with tablet servers to perform reads and writes, bypassing the master for efficiency.

• Caches tablet location data for quick access.
• Automatically recovers from stale tablet mappings.
• Uses efficient retry and backoff strategies for reliability.

The Bigtable API provides a simple interface for applications to read, write, and manage data stored in tables.

• Create/Delete Tables — Define new tables and column families or remove unused ones.
• Read/Write Operations — Access rows using Get or Scan; modify data with Put and Delete.
• Atomic Row Mutations — All operations on a single row are atomic.
• Filters — Select only specific columns, versions, or values during reads.
• Timestamps & Versions — Retrieve or write different versions of a cell by timestamp.

Bigtable’s API is intentionally simple — no joins or complex queries — focusing instead on scalable, high-throughput operations across massive datasets.