Online Boutique: Cloud-Native Microservices Demo Application

This repo is a fork of Tetrate's modified version of the GCP Hipstershop/Online Boutique with modifications to allow for in-depth Zipkin tracing. Changes include:

• Native 100% tracing of every service with Zipkin with respect to parent and child spans.
• Deployment with pre-built Zipkin and MySQL instances to allow for fast data generation and extraction.
• Rewrite of adservice, cartservice, currencyservice, emailservice, paymentservice and recommendationservice in Go. Usage of another loadgenerator.
• Lots of smaller fixes.
• Ready to use deployment and data extraction scripts. All services are deployed to separate nodes.
• Artificial delays (constant workload) with matrix multiplication for all microservices.

Overview

The following picture shows the connection graph of the services as defined by Tetrate. We reimplemented cartservice, currencyservice and paymentservice in Go. We do not build or deploy apiservice in our scripts and setup. Our adservice implementation works with grpc (again). The ingress gateway is no longer necessary to access the webpage. Here is our (slightly) updated and more detailed service architecture, which also shows the functionalities of the services:

Artificial Delays

Artificial delays can be activated by setting the DELAY_* environment variables in the .yaml deployment files in the kubernetes-manifests folder. The variables (64-bit signed) have to be set to positive integers to activate the feature. The delay variable describes the amount of matrix (random values, constant size) multiplications computed before the actual task. The table below shows some measurements from an example experiment (low utilization, on n1-standard-1 nodes):

DELAY [#matrix multiplications]	Delay [ms]	TIME / MM	Service
2	0,7	0,35	adservice
4	1,24	0,31	cartservice
6	1,63	0,271666667	cartservice
8	3,77	0,47125	currencyservice
10	2,89	0,289	currencyservice
12	1,21	0,100833333	emailservice
14	4,77	0,340714286	paymentservice
16	5,03	0,314375	productcatalogservice
18	13,42	0,745555556	productcatalogservice
20	4,95	0,2475	shippingservice
22	4,01	0,182272727	shippingservice

Building Images

Images are built automatically using a Github Action. They are published in Docker Hub in the simoneismann registry.

You can build the images using the scripts located in the hack folder:

# build only the image of emailservice
TAG=v0.1.8 REPO_PREFIX=my.docker.hub ./hack/make-docker-images-nopush.sh emailservice

# build all images locally (no push)
TAG=v0.1.8 REPO_PREFIX=my.docker.hub ./hack/make-docker-images-nopush.sh

# build all and push to Docker Registry
TAG=v0.1.8 REPO_PREFIX=my.docker.hub ./hack/make-docker-images.sh

Deployment in Google Cloud Shell

Either execute deploy_gcp_raw.sh or deploy_gcp_loadgen.sh from the repository root to deploy the system. A deprecated Istio setup script can be found in the doc folder for reference.

MySQL dump to CSV

Install a MySQL client:

sudo apt-get install mariadb-client

Then create a shell script, which takes the hostname (or IP) of the mysql host as a parameter:

HOST=$1
mkdir dump
for tb in $(mysql --protocol=tcp --host=${HOST} -pzipkin -uzipkin zipkin -sN -e "SHOW TABLES;"); do
    mysql -B --protocol=tcp --host=${HOST} -pzipkin -uzipkin zipkin -e "SELECT * FROM ${tb};" | sed "s/\"/\"\"/g;s/'/\'/;s/\t/\",\"/g;s/^/\"/;s/$/\"/;s/\n//g" > dump/${tb}.csv;
done

Measurement Scripts

We offer two measurement scripts with a complete usage of the frontend API and a checkout only workload. A few pieces of software are required to execute them:

sudo apt-get install open-jre
sudo apt-get install mariadb-client
sudo apt-get install python3
sudo apt-get install golang-go
pip3 install pandas
pip3 install openpyxl

Additionally, the firewall for all VMs should be configured to allow incoming connections on port 22442. Some usage examples are given in the shell files themselves.