Database oferings in production are generally served to clients via gateways(proxies). Instead of recreating them from scratch for each ofering, a composable architecture allows for diferent gateways to share code, providing increased reliability, engineering eficiency and better business impact. Joint Workshops at 49th International Conference on Very Large Data Bases (VLDBW'23) - Second International Workshop on Composable
At scale, database oferings are generally fronted by gateThese functionalities, when implemented correctly in ways(proxies), which acts as an intermediary between a generic modular way, can be reused and stacked toapplications and one or more database instances. At Meta gether to provide a gateway for any database ofering. for example, we have multiple databases (relational, nonBuilding a composable Gateway has major benefits in relational, graph, etc.) and each one is fronted by its own terms of developer productivity, reliability, availability gateway service. If we take a step back and squint, it is of advanced features across all oferings, reduced mainobserved that they all share common functionality. Some tenance, consistent experience for customers. Keeping of the key functions of a gateway include: • Connection Management: It provides connection pooling, abstracts and simplifies logic of establishing connections to database instances. • Authentication and Authorization: Simpli- tocols, query experiences, and one being open source and • Tenant isolation and Load balancing: ifes the responsibility of databases and ensures only valid and authorized clients can access them. this, as a first step, we are migrating MySQL to use thrift other being home grown on native technology. To resolve as the communication protocol end-to-end, bringing the These are one of the most critical pieces of a first stage of parity for both the datastores. At the same shared database ofering, where gateway is responsible for providing safety to one application or database from another noisy neighbor.
time, we are teasing out these common functionalities some of which are stated above into pluggable modules, which can be combined together into a fully composable • Routing and Discoverability: Gateway ab- gateway to serve both the oferings. This architecture stracts the layout and discoverability of backing
provides a robust, extensible and standardized solution physical DB instances from the client and pro- for accessing databases, with easier maintenance and vides routing via logical entities. • Caching: The gateways can implement caching mechanisms to store frequently accessed data, reducing the load on the databases and improving scalability at core. these functions in mind, at Meta, we have been exploring building a gateway with pluggable components for our internal Database oferings, MySQL[ 1] and ZippyDB[2].
Currently both have service specific gateways and client stacks because of how each has diferent connection proresponse times for clients. • Monitoring and Analytics: This is bread and butter, to give insights into the usage pattern for a database or application by providing a single choke point. [2] S. Masti, How we built a general purpose key value