DoorDash Uses Envoy and Valkey for a 1.5M RPS Proxy Cache with 99.99999% Availability

DoorDash’s recent announcement of Entity Cache, a transparent proxy caching platform built on Envoy and Valkey, represents a significant advancement in managing complexity within microservices architectures. The sheer scale of operation – handling 1.5 million requests per second with a staggering 99.99999% availability – underscores the importance of efficient data management at leading edge companies. This isn't just about speed; it’s about reliability and resilience in a landscape where user experience hinges on consistent performance. The challenges DoorDash faced are increasingly common as organizations embrace microservices, leading to a proliferation of service-to-service requests and potential bottlenecks. It’s fascinating to see how they’re tackling this issue, especially when considering the ongoing discussion around skill sets needed in the AI era [Am I focusing on the wrong skills as a CS student in the AI era? (Need brutally honest advice)]. The rise of AI and machine learning is influencing architectural decisions, and this caching solution feels like a pragmatic response to the demands of modern, data-intensive applications. Moreover, the emphasis on event-driven invalidation highlights a move towards more dynamic and responsive caching strategies, a far cry from traditional, static approaches.

The clever integration of Envoy and Valkey – a combination we're seeing more frequently – speaks to a broader trend of leveraging existing, robust technologies to build custom solutions. Rather than reinventing the wheel, DoorDash has intelligently combined proven components to address a specific need within their ecosystem. This contrasts with some of the hype surrounding "revolutionary" AI solutions; Entity Cache is a testament to the power of thoughtful engineering and optimization. It’s noteworthy that this platform operates within DoorDash’s service mesh, indicating a deliberate and integrated approach to infrastructure management. This mirrors the broader industry shift toward service meshes allowing developers to focus on application logic, rather than infrastructure concerns – a principle we’ve explored in discussions around deep learning tackling single-cell analysis [Deep learning tackles single-cell analysis – A survey of deep learning for scRNA-seq analysis]. The key takeaway here isn’t the specific technology stack, but the strategic approach to building a high-performance, highly available caching layer.

The 99.99999% availability figure is particularly compelling. This level of reliability requires meticulous attention to detail in failure handling and performance optimization, demonstrating a commitment to operational excellence. While many claim impressive uptime percentages, achieving this level consistently across a platform handling millions of requests demands a sophisticated architecture and rigorous monitoring. The transparency of the proxy caching platform is another critical factor; it allows developers to understand and troubleshoot performance issues more effectively, fostering a culture of continuous improvement. The value proposition extends beyond mere speed and availability; it's about empowering developers with the tools they need to build and maintain robust, scalable applications. It’s a pragmatic approach that resonates with the challenges of building world models [I just read LeCun’s recent thoughts on world models. Thoughts on JEPA as a path forward?] – focusing on the practical application of complex concepts to solve real-world problems.

Looking ahead, it will be interesting to see how other organizations adapt and adopt similar caching strategies to optimize their microservices architectures. The success of Entity Cache suggests that transparent proxy caching, built on open-source technologies like Envoy and Valkey, will become an increasingly prevalent pattern. The focus on event-driven invalidation foreshadows a future where caching is more dynamic and responsive to real-time data changes. A key question is whether we'll see more organizations publicly sharing their architectural patterns and best practices, fostering a collaborative approach to solving the challenges of scaling microservices. The rise of specialized caching layers, optimized for specific workloads, may also become a defining trend in the years to come.