June 2022 Summaries

Redis, a widely used in-memory datastore, is often compared to alternative architectures like KeyDB and Dragonfly. While these projects bring interesting technologies and ideas, Redis maintains its position as the best in class for an in-memory, real-time datastore due to its architecture principles. Redis scales horizontally by running multiple processes (using Redis Cluster) even within a single cloud instance. This allows for linear scaling, faster replication, and quick recovery from VM failure. Additionally, limiting each Redis process to a reasonable size ensures efficient memory management and easy cluster manipulation. Horizontal scaling is crucial as it provides better resiliency, easier scaling, cost-effectiveness, and high throughput. The shared-nothing, multi-process architecture of Redis offers the best performance, scalability, and resilience while supporting various deployment architectures required by an in-memory, real-time data platform.

A schemaless database allows for flexibility and adaptability without adhering to a predefined blueprint or structure, making it suitable for real-time data-driven industries such as financial services and gaming. This approach enables businesses to handle changing operations and data types without major repercussions, allowing all data and metadata to remain unaltered and accessible. In contrast, schema databases require rigorous testing, flexible data modeling, and strict adherence to rules, which can make altering the schema at a later date a laborious process. While schemaless databases offer advantages such as accommodating various data models and adding additional fields, they also lack universal compatibility with SQL instructions and ACID-level compliance, making them less suitable for certain use cases.

To mitigate data loss risks associated with Redis databases, persistence and backup mechanisms are crucial. Redis Enterprise relies on persistence for recovery when both primary and secondary database shards are lost, which can be achieved through managed backups across clouds using Cohesity SmartFiles. The integration enables easy setup of S3 compatible endpoint for backups, allowing for frictionless operational strategies. By utilizing this verified integration, users can ensure resilient and durable data platforms.

Redis is developed with a great emphasis on performance, aiming for stability and speed in every release. To monitor and analyze Redis' performance, the team runs cross-company standards for performance and observability requirements, utilizing open-source tools like memtier_benchmark and Linux perf_events. They identify potential performance regressions and improvements by running SPEC benchmarks in a "zero-touch" fully automated mode, using instruments like Brendan Greg's FlameGraph repo. Through analysis, they noticed an optimization opportunity in the Streams' ingest benchmark, which led to a 20% improvement in CPU cycles, achieved through avoiding duplicate work and non-optimal memory management within the same code block.

E-commerce sales are expected to exceed $5 trillion in 2022, accounting for more than a fifth of overall retail sales. However, only 85% of online transactions are approved compared to 97% in-store, leading to lost business opportunities. Kipp, a fintech innovator, aims to improve the payment approval process between credit issuers and merchants by leveraging Redis Enterprise Cloud on AWS for its speed, performance, and redundancy. By automating real-time collaboration between merchants and banks, Kipp's solution, Kipp Connect, delivers intelligent, customer-friendly credit decisions to reduce the number of declined transactions in e-commerce.

Redis is a key-value store that has evolved beyond its traditional purpose, and clustering allows data to be sharded across multiple nodes. In a cluster, data resides in one place, and each node or shard has a portion of the keyspace, divided into 16,384 slots. Keys are related to slots through hashing functions, which derive a single number from a string of characters. Transactions in Redis only occur within the same hash slot, ensuring high throughput and eliminating failures scenarios. To approach this with advanced planning, hashtags can be used to logically divide keys among nodes, while Redis Enterprise offers regular expressions for sharding. When designing an application for clustering, it's essential to consider command limitations and potential key issues.