Broadcast table
Broadcast Table is a table data replication policy in TDSQL Boundless, which means replicating the complete table data to all compute nodes in the cluster. This enables efficient JOIN operations with large tables or reduces distributed transactions in multi-table operations. Due to the significant overhead caused by broadcast table data synchronization, it is suitable for "write less, read more" scenarios.
Compaction
Merge/Compaction is a core mechanism in the LSM-Tree storage engine, responsible for merging multiple small data files into fewer, larger sorted files to improve query performance and optimize storage space. This mechanism performs multi-way merge sorting asynchronously in the background to clean up deleted or expired data, while reducing the number of files to minimize multi-way merge overhead during queries.
Data Affinity
Data Affinity is a technique that stores related data on the same node or nearby nodes to improve data access efficiency. By reducing the need for cross-node data transmission, it significantly lowers latency and network traffic while enhancing query performance. TDSQL Boundless adheres to automatically or manually configured data affinity rules during scheduling to ensure that data frequently accessed together is scheduled and migrated as a unit.
DDL
Data Definition Language (DDL) primarily includes commands such as CREATE, ALTER, and DROP.
DML
Data Manipulation Language (DML) commands include SELECT, UPDATE, INSERT, and DELETE.
Data Object
Data Object refers to the logical concept of database objects such as tables, partitions, and indexes.
Distributed Transaction
A distributed transaction involves collaborative processing among multiple nodes to collectively undertake a single transaction responsibility. Moreover, it externally exhibits ACID transaction properties. In TDSQL Boundless, the distributed transaction model is Two-Phase Commit + Multi-Version Concurrency Control + Time-Ordering.
Gap Lock
Gap Lock is a locking mechanism used by the MySQL InnoDB storage engine under the (REPEATABLE READ) isolation level. It locks the gaps between index records—spaces that do not define specific records but can accommodate new insertions. Its primary purpose is to prevent other transactions from inserting new data into these gaps, thereby resolving the "phantom read" issue. For instance, it may lock the gap between a range of values (such as 10 to 20), preventing the insertion of new values within this range. TDSQL Boundless also supports Gap Lock capability.
Global Timestamp
Global Timestamp is a service or mechanism in TDSQL Boundless that generates globally unique and monotonically increasing timestamp identifiers. It assigns globally consistent timestamps to distributed transactions, ensuring global transaction ordering and consistency.
Peer node
A HyperNode is a peepeer-to-peer architecture node that includes both computing and storage engines, fully supporting end-to-end features from computing to storage.
Indexes
Index is a data structure in databases used to quickly locate records that meet specific conditions. Similar to a book's table of contents, it stores key values and their corresponding data locations, allowing queries to bypass unnecessary scanning processes and significantly improve search speed.
LSM-Tree
LSM-Tree (Log-Structured Merge Tree) is a data structure specifically designed for write-intensive scenarios. By converting random writes into sequential writes, it significantly enhances write performance and is widely adopted in modern database storage engines. LSM-Tree serves as the foundational data structure for the row-based storage engine in TDSQL Boundless.
Metadata Cluster
The Meta Cluster (MC) of TDSQL Boundless is responsible for global transaction ID allocation, metadata management, and data scheduling.
MDL
Metadata Lock (MDL) is a table-level locking mechanism in TDSQL Boundless designed to ensure the consistency of database object metadata (such as table structures, indexes, and triggers, and so on) during concurrent access.
MVCC
MVCC (Multi-Version Concurrency Control) is a concurrency control mechanism in database management systems that enables non-blocking concurrent execution of read and write operations by maintaining multiple versions of data. Its core principle allows read operations to access historical data versions while write operations create new versions, thus avoiding read-write conflicts and significantly enhancing database performance in high-concurrency scenarios.
Non-Distributed Transaction
Non-Distributed Transaction is a transaction where all operations are completed on a single database node. All SQL statements in the transaction access and modify data within the same database instance, with its ACID properties guaranteed by that single node. This contrasts with distributed transactions involving multiple independent nodes. In TDSQL Boundless, transactions operating on data within the same node can be optimized as non-distributed transactions.
OLAP
On-Line Analytical Processing (OLAP) is the primary application of data warehouse systems, supporting complex analytical operations with a focus on decision support and delivering intuitively understandable query results.
Online DDL
Online Data Definition Language (Online DDL) refers to executing Data Definition Language (DDL) operations, such as modifying table structures, while keeping the database online and services uninterrupted.
OLTP
On-Line Transaction Processing (OLTP) is the primary application of traditional relational databases, focusing on fundamental day-to-day transaction processing such as bank transactions.
Optimistic Transactions
Optimistic Transaction is a transaction processing model that assumes no write conflicts will occur before execution, caching data modifications in memory and only checking for conflicts and acquiring locks at commit time. It is suitable for scenarios with fewer write operations and more read operations, offering higher performance under low conflict rates, but incurring significant rollback costs when conflicts are frequent.
Parallel query
Parallel Query (PQ) breaks down complex queries into multiple subtasks and executes them in parallel.
Partition Table
Partition Table is a database technique that divides large table data into smaller, more manageable logical units (partitions) based on specific rules. Each partition can be stored and managed independently, yet appears externally as a single complete table. In TDSQL Boundless, the partition serves as the smallest unit of scheduling.
Pessimistic Transactions
Pessimistic Transaction is a transaction processing model that preemptively locks relevant data before execution, ensuring the transaction can commit successfully. It is suitable for systems with frequent write operations, preventing write-write conflicts at the database level.
Raft algorithm
Raft Algorithm is a consensus algorithm for distributed systems that ensures data consistency among multiple nodes through leader election and log replication mechanisms. It decomposes the consensus problem into three relatively independent subproblems: leader election, log replication, and safety, making it more comprehensible and implementable compared to the Paxos algorithm. The high availability and data redundancy mechanisms of TDSQL Boundless are developed based on the Raft consensus algorithm.
Region
Region (data shard): a continuous Key space.
RPC
RPC (Remote Procedure Call) is an inter-process communication protocol that allows a program to invoke functions or procedures on another machine (or other processes on the same machine) as if calling a local function, abstracting away the complexities of network communication.
Replica
In TDSQL Boundless, data is replicated for high availability using the Raft consensus protocol, where each copy of data is called a replica (Replica). Replicas are divided into Leader and Follower roles. The Leader replica handles read and write requests, while Follower replicas primarily replicate data and may handle read requests when necessary.
Replication Group
Replication Group (RG), which contains one or more Regions, corresponds to a Raft log stream.
RocksDB
RocksDB is a high-performance embedded key-value storage engine developed by Facebook based on Google LevelDB. It adopts an LSM-Tree (Log-Structured Merge Tree) architecture, optimized for high-speed storage devices such as SSDs, supporting high write throughput and fast point queries. Widely used in distributed databases, stream processing systems, and other scenarios, TDSQL's storage engine is developed based on RocksDB.
SQLEngine
The computational engine of TDSQL Boundless is fully compatible with MySQL 8.0, responsible for SQL parsing, query optimization, and execution.
SSTable
SSTable (Sorted String Table) is an immutable data file stored on disk in the LSM-Tree, characterized by key-ordered storage. Data is written as key-value pairs sorted by primary key, enabling efficient range queries and point queries (with Bloom filters). Once written to disk, SSTables become immutable; updates and deletions are handled by appending new records, while old data is cleaned during the Compaction process.
Tables
A table is the fundamental structure for storing data in a database, consisting of rows and columns. Each row represents a record, and each column represents a field (attribute). Tables serve as the most basic building blocks of a database, organizing and storing structured data.
TDStore
The storage engine of TDSQL Boundless, based on an LSM-Tree + SSTable structure, uses the Raft consensus protocol for disaster tolerance.
2PC
2PC (Two-Phase Commit) is a classic protocol in distributed systems for ensuring transaction atomicity. Through the interaction between a coordinator and participants, it ensures that all nodes either commit the transaction entirely or roll it back entirely, achieving strong consistency.