Skip to content

RahulKumar1119/microservices-demo-1

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Online Boutique

Online Boutique is a cloud-native microservices demo application. Online Boutique consists of a 10-tier microservices application, writen in 5 different languages: Go, Java, .NET, Node, and Python. The application is a web-based e-commerce platform where users can browse items, add them to a cart, and purchase them.

Honeycomb uses this application to demonstrate use of technologies like Kubernetes, gRPC, and OpenTelemetry. This application works on any Kubernetes cluster. It’s easy to deploy with little to no configuration.

Screenshots

Home Page Checkout Screen
Screenshot of store homepage Screenshot of checkout screen

OpenTelemetry

Online Boutique is instrumented using the OpenTelemetry framework. There are simple and advanced instrumentation techniques offered by OpenTelemetry that are leveraged in the application. Each service in the src folder explains how OpenTelemetry was used with specific code examples.

Development

Prerequisites

Install

  1. Launch a local Kubernetes cluster with one of the following options:
  • To launch Minikube (tested with Ubuntu Linux). Please, ensure that the local Kubernetes cluster has at least:
    • 4 CPUs
    • 4.0 GiB memory
    • 32 GB disk space
minikube start --cpus=4 --memory 4096 --disk-size 32g
  • To launch Docker for Desktop (tested with Mac/Windows). Go to Preferences:
    • choose “Enable Kubernetes”,
    • set CPUs to at least 3, and Memory to at least 6.0 GiB
    • on the "Disk" tab, set at least 32 GB disk space
  1. Run kubectl get nodes to verify you're connected to the respective control plane.

  2. Install the OpenTelemetry Collector Helm chart from Honeycomb. Specify your Honeycomb API key when installing the Helm chart.

helm repo add honeycomb https://honeycombio.github.io/helm-charts
helm install opentelemetry-collector honeycomb/opentelemetry-collector \
      --set honeycomb.apiKey=YOUR_API_KEY \
      --values ./kubernetes-manifests/additional_resources/opentelemetry-collector-values.yaml
  1. Run skaffold run (first time will be slow, it can take ~20 minutes). This will build and deploy the application. If you need to rebuild the images automatically as you refactor the code, run skaffold dev command.

  2. Run kubectl get pods to verify the Pods are ready and running.

  3. Access the web frontend through your browser

  • Minikube requires you to run a command to access the frontend service:
minikube service frontend-external

Cleanup

If you've deployed the application with skaffold run command, you can run skaffold delete to clean up the deployed resources.

Architecture

Online Boutique is composed of 10 microservices (plus a load generator) written in 5 different languages that communicate with each other over gRPC.

Architecture of microservices

Find Protocol Buffers Descriptions at the ./pb directory.

Service Language Description
adservice Java Provides text ads based on given context words.
cartservice C# Stores the items in the user's shopping cart in Redis and retrieves it.
checkoutservice Go Retrieves user cart, prepares order and orchestrates the payment, shipping and the email notification.
currencyservice Node.js Converts one money amount to another currency. Uses real values fetched from European Central Bank. It's the highest QPS service.
emailservice Python Sends users an order confirmation email (mock).
frontend Go Exposes an HTTP server to serve the website. Does not require signup/login and generates session IDs for all users automatically.
loadgenerator Python/Locust Continuously sends requests imitating realistic user shopping flows to the frontend.
paymentservice Node.js Charges the given credit card info (mock) with the given amount and returns a transaction ID.
productcatalogservice Go Provides the list of products from a JSON file and ability to search products and get individual products.
recommendationservice Python Recommends other products based on what's given in the cart.
shippingservice Go Gives shipping cost estimates based on the shopping cart. Ships items to the given address (mock)

Features

  • Kubernetes: The app is designed to run on Kubernetes
  • gRPC: Microservices use a high volume of gRPC calls to communicate to each other.
  • OpenTelemetry Tracing: Most services are instrumented using OpenTelemetry trace providers for gRPC/HTTP.
  • Skaffold: Application is deployed to Kubernetes with a single command using Skaffold.
  • Synthetic Load Generation: The application demo comes with a background job that creates realistic usage patterns on the website using Locust load generator.

History

This project originated from the excellent Google Cloud Platform Microservices Demo. It was forked in 2021, before significant changes were performed. All application telemetry which was previously done with OpenCensus and Stackdriver, was moved to use OpenTelemetry for application telemetry, with tracing export intended for an OpenTelemetry Collector. Additional instrumentation is leveraged throughout the application to show some basic and advanced capabilities of OpenTelemetry. This application is used as a demo platform for the Honeycomb team, and many changes were made to the application code so it will break in ways that make for a more compelling demonstration of the Honeycomb platform.

Application demo

This application will exhibit a problem meant to be discovered with ease using the Honeycomb platform.

The checkout service has a memory leak, cause by an internal cache store. This service has tight Kubernetes pod/container memory limits, so the leak will cause out of memory crashes, resulting in a pod restart after approximately 4 hours. Code in the checkout service will introduce additional delays in the form of SQL calls under getDiscounts. The number of SQL calls made will increase as the cache size increases, creating exponentially increasing latency. There is additional code in the frontend service, which will introduce a specific userid (20109) after the cache limit from checkout has reached a specific threshold. This results in a pattern where a single user from a pool of thousands, receiving a bad experience that continues to get worse.

When using Honeycomb BubbleUp, and combined with the Honeycomb SLO feature, understanding the single user from the high cardinality pool of thousands of user ids is easy to do. Honeycomb allows the user to ask novel questions from the data, to quickly understand the memory leak and cache problem in code.


This is not an official Honeycomb or Google project.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published