Managed Transactional Persistence

事务是基本的、无处不在的，是各种数据密集型应用程序的关键基础：从小型嵌入式系统到全球、始终在线的消费者和企业应用程序。事务可以被认为是应用程序和支持它们的底层数据系统之间的契约。 然后应用程序依赖此契约来确保正确的操作。 持久性是事务契约的重要组成部分。 非正式地说，持久性是保证失败不会删除应用程序事务提交的数据库更改。 自从40多年前事务首次正式化以来，持久性契约几乎没有什么变化。 不幸的是，它已经越来越不符合现代数据系统和应用程序的现实。 这种不匹配给应用程序带来了真实的问题，应用程序必须退回到特定的技术，而这些技术破坏了事务契约应该提供的许多优势。 我建议开发一种新的事务数据系统，可以解决持久性问题。 由于问题的根源在于持久性契约本身，因此这项工作的一个重要贡献将是一个新的、灵活的持久性契约。 新合同将允许应用程序直接管理性能和数据丢失风险之间的权衡，而无需诉诸当前的临时技术。 我将演示这个新的持久性契约可以用作在各种设置中构建更好的应用程序和数据系统的基础。

Transactions are fundamental and ubiquitous, a critical foundation for data intensive applications of all kinds: from tiny embedded systems to global, always-on, consumer and enterprise applications. Transactions can be thought of as a contract between applications and the underlying data system that supports them. Applications then rely on this contract to ensure correct operation. Persistence is an important part of the transactional contract. Informally, persistence is a guarantee that failures will not erase database changes that have been committed by an application's transactions. The persistence contract has changed little since transactions were first formalized more than 40 years ago. Unfortunately, it has grown increasingly out of step with the realities of modern data systems and applications. This mismatch causes real problems for applications, which must fall back to ad hoc techniques that destroy many of the advantages that the transactional contract ought to provide. I propose to develop a new breed of transactional data systems that can solve the persistence problem. Since the root of the problem is the persistence contract itself, an important contribution of this work will be a new, flexible persistence contract. The new contract will allow applications to directly manage a tradeoff between performance and the risk of data loss, without resorting to current ad hoc techniques. I will demonstrate that this new persistence contract can be used as a foundation for building better applications and data systems in a wide variety of settings.