Spanner中的（TrueTime，Lock free snapshot transactions，Nonblocking reads，Atomic schema changes）
Spanner 是 Google 的可扩展、多版本支持、全球分布式的同步备份数据库
TrueTime API
The TrueTime API of Spanner is designed to address the issue of time uncertainty in distributed systems. It is based on the following key concepts:
- Multi-copy replication: Spanner uses multi-copy replication mechanism to ensure data availability and consistency. It replicates data to multiple geographical locations and ensures data durability and redundancy through asynchronous replication.
- Distributed scheduler: The scheduling strategy in Spanner is distributed and takes into account the notion of time. The scheduler determines the order of writes among replicas and ensures linearizability of data writes. This enables reliable time ordering and consistency in a distributed system.
- TrueTime API: The TrueTime API is a core component of Spanner that provides an accurate and reliable timestamp service. It is based on physical time references and information from GPS satellites to provide a monotonic and global time.
- Physical time references: Spanner uses physical time references to provide accurate time. Physical time references typically come from external sources, such as GPS satellites. By using physical time references, Spanner can provide a globally increasing and ordered time sequence.
- Clock synchronization: In Spanner, clock synchronization is crucial. Each node in Spanner regularly calibrates its internal clock with the physical time. This ensures that Spanner provides accurate time information throughout the system.
Through these design concepts, the TrueTime API can provide a trusted and accurate time service for handling time uncertainty in distributed systems. Spanner leverages these design concepts to tightly control time and provide distributed data processing capabilities with strong consistency and reliability.
Spanner的TrueTime API是为了解决分布式系统中的时间不确定性而设计的。它的设计思想是基于以下几个关键概念：
- 多副本复制：Spanner使用多副本复制机制来确保数据的可用性和一致性。它将数据复制到多个地理位置，并通过异步复制来确保数据的持久性和冗余。
- 分布式调度器：Spanner中的调度策略是分布式的，并且考虑了时序的因素。调度器负责确定副本之间的写入顺序，并确保数据写入的线性一致性。这样可以在分布式系统中实现可靠的时间排序和一致性。
- TrueTime API：TrueTime API是Spanner的核心组件，它提供了一种准确和可靠的时间戳服务。TrueTime API是基于物理时间基准和GPS卫星的信息来提供一个单调和全局的时间。
- 物理时间基准：Spanner使用物理时间基准来提供准确的时间。物理时间基准通常来源于外部时间源，如GPS卫星。通过使用物理时间基准，Spanner可以提供一个全局递增的和有序的时间序列。
- 时钟校准：在Spanner中，时钟校准是非常重要的。Spanner中的每个节点都会定期校准自己的内部时钟和物理时间。这样可以保证Spanner在整个系统中提供准确的时间信息。
通过这些设计思想，TrueTime API可以提供一个可信赖且准确的时间服务，用于处理分布式系统中的时间不确定性问题。Spanner利用这些设计思想，对时间进行严格控制，并提供了强一致性和可靠性的分布式数据处理能力。
Nonblocking reads in the past
In Spanner, nonblocking reads are achieved through a combination of distributed and replicated data storage, as well as a distributed transaction protocol, to ensure high availability and consistency.
To enable nonblocking reads, Spanner utilizes two key technologies: TrueTime and the distributed consistency protocol.
- TrueTime: Spanner leverages TrueTime, an internal technology developed by Google, to achieve globally consistent timestamps. TrueTime combines physical clocks and GPS signals to provide highly accurate timestamps that maintain global ordering of data across different geographical locations.
- Distributed Consistency Protocol: Spanner employs a distributed consistency protocol called “Percolator.” This protocol builds on the Chubby lock service and the Sawzall scripting language, both developed by Google. The Percolator protocol allows read operations in Spanner to proceed without blocking other operations. When a read operation occurs, it uses TrueTime to determine the most recent replica available for reading, without blocking other transactions that may be modifying or accessing the data.
By leveraging these technologies, Spanner enables nonblocking reads. Multiple read operations can occur concurrently, ensuring data consistency without impacting other ongoing transactions or operations.
In summary, Spanner achieves nonblocking reads through the use of TrueTime and a distributed consistency protocol. This allows for concurrent read operations, maintaining high availability and consistency while improving system performance and efficiency.
Spanner实现非阻塞读取的方式是通过分布式和复制的数据存储结构以及分布式事务协议。这些技术确保了数据的高可用性和一致性，并使得多个读操作可以同时进行，而不会互相阻塞。
具体来说，Spanner采用了两个关键的技术：TrueTime和分布式一致性协议。
- TrueTime：Spanner使用Google内部开发的TrueTime技术来实现全局一致的时间戳。TrueTime通过集成物理时钟和GPS信号，为分布式系统提供准确的时间戳。这使得Spanner能够在不同的地理位置之间维护数据的全局有序性。
- 分布式一致性协议：Spanner使用了一种称为"Percolator"的分布式一致性协议。该协议基于Google的Chubby锁服务和Google内部的Sawzall脚本语言。Percolator协议允许Spanner的读操作在不阻塞其他操作的情况下进行。当一个读操作发生时，它会使用TrueTime来确定允许读取的最新副本，而不会阻塞其他事务对数据的修改或访问。
通过这些技术的支持，Spanner可以实现非阻塞读取。读操作可以并发地进行，同时保证数据的一致性，而不会影响其他正在进行的事务或操作。
总结而言，Spanner的非阻塞读取利用了TrueTime和分布式一致性协议，在实现高可用性和一致性的同时，允许并发的读操作，从而提高了系统的性能和效率。
Lock free snapshot transactions
In Spanner, a lock-free snapshot transaction is a transaction processing technique that allows for concurrent execution of transactions without the need for locks. Spanner is a distributed database system developed by Google, known for its globally consistent and highly scalable architecture.
The design philosophy behind lock-free snapshot transactions in Spanner revolves around two key concepts: TrueTime and Multi-Version Concurrency Control (MVCC).
Firstly, TrueTime is a time service in Spanner that provides a globally synchronized time for all nodes in the system. It ensures that all nodes have a consistent and ordered view of time, which is crucial for enforcing transaction consistency and repeatability across distributed systems.
Secondly, MVCC is the concurrency control mechanism employed by Spanner. Each data object in Spanner has multiple versions, each associated with a timestamp. When a transaction reads data, it queries the system for the most recent version of the data that predates its own start timestamp, ensuring a consistent snapshot. This versioning scheme enables concurrent transactions to read from the database without blocking each other. Conflicts between transactions are resolved by Spanner’s coordinator component, allowing for mutually exclusive write operations.
By combining TrueTime and MVCC, Spanner achieves lock-free snapshot transactions. Transactions can execute concurrently without the need for locks, and they can read consistent snapshots without interference from other transactions. This design philosophy and implementation principle allow Spanner to provide strong consistency and high concurrency, making it a reliable distributed database solution.
在Google Spanner中，无锁快照事务是一种关键的设计思想和实现原理。Spanner是Google开发的高度可扩展的分布式数据库系统，它提供了全球范围的数据一致性和事务支持。
Spanner通过使用两个重要的机制来实现无锁快照事务：TrueTime和多版本并发控制（MVCC）。
首先，TrueTime是Spanner中的时间服务，用于全局同步所有节点的时间。它提供了一个严格有序的时间序列，并确保各个节点之间的时间协调。这有助于事务的一致性和可重复性。
其次，MVCC是Spanner的并发控制机制。每个数据对象在Spanner中都有多个版本，每个版本都具有与其关联的时间戳。事务在读取数据时，会向系统查询从TrueTime获得的时间戳，以确保读取的是先于事务开始时间的最新版本。这种版本控制允许并发事务同时进行读取操作，而不会相互阻塞。事务之间的冲突由Spanner的协调器组件进行处理，确保事务以互斥的方式进行写入操作。
通过TrueTime和MVCC的组合，Spanner实现了高度并发的无锁快照事务。事务可以在跨多个节点和数据中心的分布式系统中执行，而不会受到锁机制的限制。它可以读取一致的数据快照，而不会受到其他事务的干扰。这种设计思想和实现原理使得Spanner能够提供强一致性和高并发性能，使其成为一种可靠的分布式数据库解决方案。
Atomic schema changes
In Spanner, atomic schema changes refer to the design principle and implementation mechanism that allows for seamless modification of the database schema without disrupting existing data. It enables structural changes such as adding or deleting tables, modifying table columns, changing indexes, etc., without interrupting the normal operation of running applications.
Spanner achieves atomic schema changes through the following approaches:
- Transactional operations: Spanner utilizes distributed transactions to achieve atomic schema changes. When performing schema modifications, Spanner treats all related operations as a single atomic transaction. This ensures that either all changes are applied or no changes are applied. In case of failures or aborts, the system can roll back the changes and maintain the consistency of the database.
- Data replay: Spanner employs data replay techniques to accomplish atomic schema changes. During the schema modification process, Spanner records the change operations in logs and then replays these operations to effect the schema changes. The logging and replay mechanism guarantee that all changes are consistently applied to the database.
- Distributed architecture: The distributed architecture of Spanner plays a crucial role in implementing atomic schema changes. Spanner divides the database into multiple partitions and distributes these partitions across multiple physical nodes. When performing schema modifications, Spanner can execute change operations in parallel across the nodes, thereby improving efficiency and performance.
To summarize, Spanner achieves atomic schema changes through the use of distributed transactions, data replay, and distributed architecture. This design principle and implementation mechanism ensure the consistency and reliability of schema modifications, allowing the database to undergo structural changes without interrupting the normal operation of applications.
在Spanner中，atomic schema changes（原子模式更改）是一种设计思想和实现原理，旨在允许数据库的模式（schema）在不中断现有数据的情况下进行更改。这意味着数据库可以进行结构上的变更，如添加或删除表、修改表的列、更改索引等，而不会中断正在运行的应用程序的正常操作。
Spanner实现了原子模式更改的目标，可以通过以下方式进行：
- 事务性操作：Spanner使用分布式事务来实现原子模式更改。在进行模式更改时，Spanner会将所有相关操作作为一个原子事务来执行，确保要么所有更改都被应用，要么没有更改被应用。如果发生故障或者中止，系统可以回滚更改并保持数据库的一致性状态。
- 数据重放：Spanner使用数据重放技术来实现原子模式更改。在进行模式更改时，Spanner会将更改操作记录在日志中，然后重新播放这些操作以实现模式更改。日志记录和重放机制保证了所有更改都以一致的方式应用到数据库中。
- 分布式架构：Spanner的分布式架构对于实现原子模式更改也起到了重要作用。Spanner将数据库分为多个分区，并将这些分区分布在多个物理节点上。当进行模式更改时，Spanner可以在节点之间并行执行更改操作，以提高效率和性能。
总结起来，Spanner通过使用分布式事务、数据重放和分布式架构来实现原子模式更改。这种设计思想和实现原理确保了模式更改的一致性和可靠性，使得数据库可以在不中断应用程序的情况下进行结构上的变更。
引用
LLMs
https://dl.acm.org/doi/pdf/10.1145/2491245