费良宏 Technical Evangelist | Amazon Web Services lianghon@amazon.com 用 Python 开发 Serverless 应用 @ 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved $whoami • 软件工程师 (1991 – 今) • （2011-1025） • Architect Evangelist @ (2004-2011) … Serverless 为何而生 ？ 为什么需要“无服务器”？ 服务部署及利用率 可用性及容错度 监控、调度 运行及管理 什么是“Serverless”? 无需管理基础设施 自动扩展 按价值付费 高可用以及安全 “无服务”是一种新的模型，它跨越许多不同类别的服务 AWS API Gateway Amazon SNS Amazon SQS AWS Step Functions 计 算 数 据 存 储 集 成 Amazon Aurora Serverless Amazon S3 Amazon DynamoDB AWS AppSync 2014.11 2016.2 2016.3 2017.12 AWS Lambda Google Cloud

A Little bit about Myself • Serverless Specialist SA, GCR • Based in Taipei • Focus on Serverless, Container and AWS CDK • Father of a daughter • Speaker in AWS Global Summits, KubeCon and CNCF Webinar @pahudnet hunhsieh Pahud Hsieh 目录 CONTENTS Why Serverless How Serverless in Python Packaging and Distribution Serverless to the Next Level Why Serverless Development transformation at Amazon: 2001–2002 No Lambda functions Start End Serverless compute engine for containers Long-running Bring existing code Fully-managed orchestration AWS Fargate Serverless event-driven code execution Short-lived

Cloud-Native PaaS for microservices and serverless Cloud Infrastructure-Ready VMs Managed services Azure PaaS for containerized microservices + Serverless computing + Managed services And Windows Identity/Auth, Cost control and Operations model Azure Azure 粘合剂 – Azure Functions Serverless in Azure – FaaS 是 Serverless 架构的核心 什么是 Azure Functions .NET, Node, Java, Docker, PHP, Ruby, Python Staging

例子：生成测试用户 例子：远程控制饮水机 接下来： • 支持更多的数据类型 • 类型对应的控件可配置 大胆的想法： • 直接使用 Function 构建小型应用的管理后台？ • Serverless 管理后台？ 谢谢大家

Infrastructure as Code Infrastructure is Code vs AWS Cloud Development Kit (CDK) 虚拟机 容器 无服务器 Serverless 数据库 缓存服务 Hadoop 容器编排 微服务 Fargate AWS Lambda Hello World • 事件响应 一个函数对应一类事件 • API后端 一个函数对应一个API

How to Zoom a Clip to a GUI Window Please also read the previous section. This time we want to compute the zoom factor for a clip, such that its image best fits a given GUI window. This means, that the get_pixmap(matrix=mat, clip=clip) For the other way round, now assume you have the zoom factor and need to compute the fitting clip. In this case we have zoom = HEIGHT/clip.height = WIDTH/clip.width, so we must HEIGHT/zoom and, clip.width = WIDTH/zoom. Choose the top-left point tl of the clip on the page to compute the right pixmap: width = WIDTH / zoom height = HEIGHT / zoom clip = fitz.Rect(tl, tl.x + width

automatically when the request is finished, but may be called earlier for applications that override compute_etag and want to do an early check for If-None-Match before completing the request. The Etag header 3.2.2: Added support for cookie version 2. Both versions 1 and 2 are supported. RequestHandler.compute_etag() Computes the etag header to be used for this request. By default uses a hash of the content RequestHandler.set_etag_header() Sets the response’s Etag header using self.compute_etag(). Note: no header will be set if compute_etag() returns None. This method is called automatically when the request

automatically when the request is finished, but may be called earlier for applications that override compute_etag and want to do an early check for If-None-Match before completing the request. The Etag header 3.2.2: Added support for cookie version 2. Both versions 1 and 2 are supported. RequestHandler.compute_etag() Computes the etag header to be used for this request. By default uses a hash of the content RequestHandler.set_etag_header() Sets the response’s Etag header using self.compute_etag(). Note: no header will be set if compute_etag() returns None. This method is called automatically when the request

BFMMLA • 128 bit 向量寄存器 • 单指令完成 (2x4) * (4x2) • 16 mul + 16 add 深度学习推理加速 • BF16 gemm 实现 • ARM Compute Library • OpenBLAS • TensorFlow • oneDNN + ACL • DNNL_DEFAULT_FPMATH_MODE=BF16 • PyTorch • OpenBLAS • oneDNN + ACL • torch.set_float32_fast_math_mode("BF16") 深度学习推理加速 • BF16 gemm 实现 • ARM Compute Library • OpenBLAS • TensorFlow • oneDNN + ACL • DNNL_DEFAULT_FPMATH_MODE=BF16 • PyTorch • OpenBLAS • oneDNN + ACL • torch.set_float32_fast_math_mode("BF16") 深度学习推理加速 • BF16 gemm 实现 • ARM Compute Library • OpenBLAS • TensorFlow • oneDNN + ACL • DNNL_DEFAULT_FPMATH_MODE=BF16 • PyTorch •

automatically when the request is finished, but may be called earlier for applications that override compute_etag and want to do an early check for If-None-Match before completing the request. The Etag header 3.2.2: Added support for cookie version 2. Both versions 1 and 2 are supported. RequestHandler.compute_etag() Computes the etag header to be used for this request. By default uses a hash of the content RequestHandler.set_etag_header() Sets the response’s Etag header using self.compute_etag(). Note: no header will be set if compute_etag() returns None. This method is called automatically when the request