#serverless
The author, Liu Yu, has a black book cover.
Finished very quickly and wrote several demos in Python.
Three parts:
- Concepts and Products
- Definitions, specifications, advantages, challenges, open source recommendations
- Getting Started with Development
- Building FaaS (Function as a Service)
- Engineering Practices
- Operations, images, audio/video, AI, frontend, etc.
In 2018, Google released Knative, an open-source serverless framework based on Kubernetes, with the ability to build containers, traffic routing, elastic scaling, zero instances, function events, etc.
CNCF's CNCF Serverless Whitepaper 1.0.
In China, there is KubeCon + CloudNativeCon.
Serverless = FaaS + BaaS (Backend as a Service).
FaaS solution consists of event source, FaaS controller, function instance, platform services (other cloud services such as DB, auth, etc.).
Cold start:
- Instance reuse. Retain for a period of time after the end of the lifecycle.
- Instance warm-up
- Passive
- Rules, algorithms, hybrid warm-up
- Weighting, deep learning predictions, etc.
- Active
- Simple configuration, metric configuration
- Hybrid
- Passive
- Resource pooling
AWS Lambda, Alibaba Cloud Function Compute, Huawei Cloud Function Workflow, Tencent Cloud Cloud Function.
CNCF provides some industrial FaaS platforms, such as Knative, OpenWhisk, Fission, Kubeless.
Open-source frameworks: Serverless Devs, Serverless Framework.
Function reuse. Perform operations before the main function.
- In machine learning scenarios, load the model during initialization to avoid reloading it every time it is triggered.
- Link objects during initialization, such as pooling.
- Download initialization content for the first run.