Why Dapr Extends Service Mesh: A Deep Dive into Multi‑Runtime Architecture

Dapr, an open‑source distributed runtime launched by Microsoft in October 2019, reached its v1.0 release in February 2021 and quickly gained traction, earning over 12 000 GitHub stars. Alibaba Cloud has been an early adopter and major contributor, with dozens of internal applications using Dapr and two Dapr members on the project, making it the largest non‑Microsoft contributor.

The article first revisits the definition of Service Mesh as an infrastructure layer that handles inter‑service communication, typically implemented via lightweight sidecar proxies that are transparent to applications. Key characteristics include positioning as infrastructure, handling service‑to‑service traffic, sidecar deployment, and non‑intrusiveness.

It then explains the sidecar pattern: unlike traditional RPC frameworks, Service Mesh moves communication logic into sidecar proxies controlled by a control plane, simplifying application code. A historical view of Istio’s evolution (protocol support, VM integration, observability, external integration) illustrates broader Service Mesh trends.

Building on this, the article introduces the Multi‑Runtime concept proposed by Bilgin Ibryam in 2020. Multi‑Runtime abstracts distributed capabilities as APIs rather than protocol‑specific proxies, allowing multiple runtimes (Mecha) to provide the same capabilities. The two‑step process involves extracting distributed needs into sidecars/proxies and then consolidating them into a minimal set of runtimes.

Key differences between Multi‑Runtime and traditional Service Mesh are highlighted: broader capability scope (beyond point‑to‑point communication), flexible deployment models (sidecar, node, edge), and API‑centric interaction with applications, enabling language‑agnostic SDKs.

Applying Multi‑Runtime to Dapr, the article details Dapr’s architecture: a sidecar runtime exposing HTTP and gRPC APIs, a set of Building Blocks (state management, pub‑sub, bindings, etc.), and Components that implement each block. Dapr’s control plane consists of Actor Placement, Sidecar Injector, Sentry, and Operator services, currently deployed as microservices.

Dapr’s vision—"Any language, any framework, anywhere"—is supported by multi‑language SDKs and a pluggable component model with over 70 implementations, enabling deployment across public, private, and edge clouds.

The article also chronicles Dapr’s milestones (0.1.0 in Oct 2019, v1.0 in Feb 2021) and Alibaba’s involvement: early internal trials, code contributions, and plans for deeper engagement.

Finally, the author looks ahead to the future of cloud‑native applications, arguing that sidecar‑based runtimes like Dapr will drive multi‑language support, portability, and lightweight microservices, while the industry moves toward serverless, multi‑cloud, and hybrid‑cloud deployments.