2021 GIAC Cloud Native Conference Highlights: Service Mesh, SkyWalking, Dubbogo 3.0

1. Controlling the Explosion Radius of Cloud‑Native Services

The opening talk by Amazon senior engineer Huang Shuai described a historic AWS outage caused by a configuration‑center failure, illustrating how strict CAP theorem constraints can lead to cascading failures. To mitigate such "explosion radius" risks, the speaker recommends isolation techniques:

Appropriate service granularity : Avoid overly fine‑grained microservices that increase operational complexity and deployment cost.

Full isolation : Deploy related services in different power zones and network segments ("not near" and "not far") to reduce correlated failures.

Random partitioning : Distribute requests across multiple queues or cells, similar to Kubernetes APF shuffle sharding, to prevent single‑point overload.

Chaos engineering : Continuously inject failures to discover and eliminate weak points before they occur.

2. Monitoring Kubernetes Events with SkyWalking

SkyWalking, traditionally a microservice observability platform, introduced the skywalking-kubernetes-event-exporter to collect, filter, and forward Kubernetes events to its backend. This enables full‑stack visibility of pod scheduling, resource allocation, and other lifecycle events.

3. Kuaishou Service Mesh Practice

Jiang Tao from Kuaishou presented the evolution of their Service Mesh across three generations, positioning Dapr as a third‑generation solution. Their architecture emphasizes moving core capabilities to the data plane and standardizing request protocols.

Unified operations improve observability, stability, fault injection, and traffic recording.

Extended Envoy to transmit only changed data, reducing per‑instance service count.

Extensive protocol‑stack and serialization optimizations.

Failure‑oriented design allows Mesh to fall back to direct connections.

Key challenges identified:

Cost: Unified deployment and operation in complex environments.

Complexity: High scale, performance demands, and intricate policies.

Adoption: Limited business‑side demand, with performance overhead concerns.

To drive adoption, Kuaishou focuses on ensuring system stability first, leveraging major company projects, extending via WASM, and establishing flagship use cases.

4. Dubbogo 3.0: Dubbo in the Cloud‑Native Era

The speaker, also the Dubbogo community lead, highlighted two major themes: Service Mesh forms and flexible services. Dubbogo 3.0 introduces a proxy‑less Service Mesh, leveraging gRPC concepts for zero‑performance loss, while acknowledging gRPC's multi‑language limitations.

To address Dubbo's language gaps, the dubbo-go-pixiu project provides gateway and sidecar modes, unifying north‑south and east‑west traffic.

Flexible services implement adaptive throttling and load balancing to cope with heterogeneous machine specs, evolving service topologies, traffic spikes, and variable upstream capacities.

Adaptive throttling : Calculates queue_size = limit * (1 - rt_noload/rt) per method, rejecting requests when inflight exceeds the computed limit and returning the deficit to the client.

Adaptive load balancing : Clients receive the server‑computed queue size and use weight‑based or P2C algorithms to select instances, preventing herd effects.

Conclusion

The author reflects on attending numerous technical conferences since 2017, noting a 9.65 overall rating for this GIAC Cloud Native track. The recap captures the evolving cloud‑native landscape, practical lessons from industry leaders, and emerging directions for Service Mesh and Dubbo in hybrid‑cloud environments.