Service Mesh
最后更新于
最后更新于
A Service Mesh is a squadron of network proxies designed to ensure secure, fast, and reliable communication between services. This layer of infrastructure came to the fore with the rise of microservices and cloud-native applications. Often deployed alongside applications in a lightweight network proxy (for instance, using the sidecar method as diagrammed below), the Service Mesh could be envisioned as a networking model superior to TCP/IP that abstracts the mechanisms for reliable inter-service communication. However, unlike TCP, it is oriented towards the applications themselves, offering unified visualization and control over them.
To ensure reliable communication between services, a Service Mesh needs to back several functions such as circuit-breaker mechanisms, latency-aware load balancing, service discovery, retries, and more. Take Linkerd, for instance; when processing a request, their workflow includes:
Determining the service for the request via dynamic route lookups
Locating instances of that service
Choosing the most optimal instance based on factors like response latency
Forwarding the request to the optimal instance, and noting latency and responses
On failure of the request or the instance becoming ineffective, forwarding to another instance for retry (provided the request is idempotent)
Failing directly if the request exceeds its time limit, thereby not increasing the load on the backend
Recording details of the request's metrics and distributed tracing
So, why exactly is a Service Mesh necessary?
It decouples service governance from the actual services, preventing intrusion into applications during the process of micro-services.
It accelerates the transformation of traditional applications into microservices or cloud-native applications.
Far from being a completely new feature, the Service Mesh merely separates functions that already existed in many applications, bringing them under the control of a unified component. Especially in the realm of microservice applications, where the number of services may be astronomical, and possibly built on various frameworks and languages, a separate Service Mesh component simplifies management and coordination.
Some popular Service Mesh frameworks include:
