jeperez / spring-cloud-microservices

Experiments in spring cloud.

spring-cloud-microservices

Experiments in spring cloud and the Netflix OSS stack

Areas to experiment in

• Hystrix (DONE)
• Zuul (DONE)
• Feign (DONE)
• Eureka (DONE)/Consul
• Vault
• Dynamic configuration
• ELK (DONE)

Setup

Pre-requisites

• Java 1.8
• Docker
• Docker-compose

Installation (debian base)

Install Docker

apt-get install apparmor lxc cgroup-lite
wget -qO- https://get.docker.com/ | sh
sudo usermod -aG docker YourUserNameHere
sudo service docker restart

Install Docker-compose (1.6+)

MAKE SURE YOU HAVE AN UP TO DATE VERSION OF DOCKER COMPOSE

To check the version:

docker-compose --version

To install the 1.6.0:

sudo su
curl -L https://github.com/docker/compose/releases/download/1.6.0/docker-compose-`uname -s`-`uname -m` > /usr/local/bin/docker-compose
chmod +x /usr/local/bin/docker-compose
exit

Install supporting tools

sudo apt-get install jq apache2-utils jmeter

To run

Start the ELK service which receives all docker logs

./ms run_elk

Execute the following to run the services.

./ms build
./ms run

Note, there are a fair number of services, mostly java, and as such they have a reasonably hefty memory requirement on aggregate.

Run ./ms redis_load_data to populate test data. The needs to be run only once.

Gradle daemon

To accelerate local development, it is recommended to run gradle daemonized. This is as simple as running the following

echo "org.gradle.daemon=true" >> ~/.gradle/gradle.properties

To test Hystrix circuit breaking

1. Build and run the system ./ms build && ./ms run

2. Check memory and uptime via ./ms stats. When the system is stable and all services have started CPU usage on all processes should be nominal.

3. Generate some load ab -n 100000 -c 13 -l http://localhost:8080/composite/[async/]2. There is also a simple JMeter load script under \configuration

4. Adjust performance of underlying services

   curl "localhost:8080/person/set-processing-time?minMs=1000&maxMs=2000" | jq .

   curl "localhost:8080/person_rec/set-processing-time?minMs=1000&maxMs=2000" | jq .

   curl "localhost:8080/product_rec/set-processing-time?minMs=1000&maxMs=2000" | jq .

   or adjust service error rate

   curl "localhost:8080/[person|person_rec|product_rec]/set-error?percentage=20" | jq .

   or disable a service

   docker-compose [un]pause [person-service|person-recommendation-service|product-recommendation-service]

5. Observe Hystrix dashboard for impact ./ms portals (shows eureka, hystrix & turbine dashboards)

Grafana setup

Hystrix and JVM metrics can be gathered for instrumentation trending and displayed via the following steps

1. Navigate to localhost:80 in a browser. Login is admin \ admin
2. Create a new data source for graphite. The url is http://localhost:81. Verify connectivity.
3. Import the two dashboards found under the \configuration\grafana directory. Remember to update the time period appropriately.

While Hystrix collects metrics which can be pushed via standard means to statsd and other solutions, Turbine does not appear to support the same metrics gathering at this time meaning metrics from all services need to be pushed and centrally aggregated in statsd rather than keys off Turbine which would be preferred.

Hystrix notes

• Thread pools are defined by group not by command. Threadpool bulkheads should be considered in the context of sequential call protection vs concurrent
• Thread pool saturation must be handled outside of Hystrix. By it's fundamental nature, if there is not a thread in the pool available Hystrix by definition cannot handle it, even if only to use a fallback.

Turbine - Hystrix Stream aggregation

• While turbine v1 support property based configuration, turbine v2 only supports discovery explicitly via Eureka. Patches seem slow to be propagated by Netflix team.
• Spring Cloud Turbine will support Consul on next release
• Eureka client versions referenced in 1.0.6 are incompatible with the Brixton.M4 Eureka service and should be excluded.

ELK Notes

• ELK is available to analyze the main spring boot application logs
• This is achieved via the docker GELF log driver in a trivially simple manner

ELK CAVEATS

• Solution requires communication through docker host
• Solution requires ELK to be available before dependent containers are started
• GELF is not secure on the wire
• GELF is less problematic using UDP and as such is likely to drop messages
• Current setup is lossy. If ELK is down messages will be lost

Eureka Notes

• Clients are now auto-deregistered upon shutdown
• Clients connect to Eureka in bootstrap phase of spring-cloud app as as such need to be identified correctly.
• The bootstrap phase provides a convenient hook to centralized configuration management services such as Vault
• Query services curl -s -H "Accept: application/json" http://localhost:8761/eureka/apps | jq '.applications.application[]'

Service overview

Service	Port	Description
person-service	-	Basic micro service with redis backend
person-recommendation-service	-	Micro service stub
product-recommendation-service	-	Micro service stub
person-composite-service	-	Composite service calling above 3 microservices
zuul edge service	8080	Edge server proxy to microservices
eureka	8001	Service discovery service
hystrix-dashboard	8000	SSE visualization tool (Hystrix specific)
turbine	8002/3	SSE aggregation service
redis	6379	Backing person service persistence
scope	4040	Docker visualization

References

Spring Boot Cloud/Netflix OSS

• https://netflix.github.io/
• http://cloud.spring.io/spring-cloud-netflix/spring-cloud-netflix.html

Service discovery

• http://technologyconversations.com/2015/09/08/service-discovery-zookeeper-vs-etcd-vs-consul/
• http://cloud.spring.io/spring-cloud-consul/#quick-start
• http://gliderlabs.com/registrator/latest/user/quickstart/

Hystrix

• https://github.com/Netflix/Hystrix/wiki/Configuration
• https://ahus1.github.io/hystrix-examples/manual.html
• https://github.com/Xorlev/crankshaftd
• http://1.bp.blogspot.com/-JaBO-Dk9oNI/VMzVWnxeP_I/AAAAAAAABOA/bYBwau60AVw/s1600/hystrix-grafana.png

Instrumentation

• https://objectpartners.com/2015/05/07/intelligent-microservice-metrics-with-spring-boot-and-statsd/
• https://github.com/ryantenney/metrics-spring

Docker host Monitoring

• https://www.brianchristner.io/how-to-setup-docker-monitoring/
• https://www.brianchristner.io/how-to-setup-prometheus-docker-monitoring/
• https://github.com/vegasbrianc/prometheus

Load testing

• http://dak1n1.com/blog/14-http-load-generate/
• http://gatling.io/#/
• https://blazemeter.com/blog/open-source-load-testing-tools-which-one-should-you-use

ELK

• http://elk-docker.readthedocs.org/#running-with-docker-compose
• http://nathanleclaire.com/blog/2015/04/27/automating-docker-logging-elasticsearch-logstash-kibana-and-logspout/
• https://github.com/gliderlabs/logspout
• http://www.labouisse.com/how-to/2015/09/14/elk-and-docker-1-8/
• http://www.labouisse.com/how-to/2015/09/23/elk-docker-and-spring-boot/
• http://knes1.github.io/blog/2015/2015-08-16-manage-spring-boot-logs-with-elasticsearch-kibana-and-logstash.html

Undertow vs Tomcat

• http://www.alexecollins.com/spring-boot-performance/

Vault

• https://github.com/markramach/spring-boot-starter-vault

Smaller Java 8 docker images & memory

• https://developer.atlassian.com/blog/2015/08/minimal-java-docker-containers/
• https://github.com/cloudimmunity/docker-slim
• https://spring.io/blog/2015/12/10/spring-boot-memory-performance
• https://blog.takipi.com/lean-mean-java-virtual-machine-making-your-docker-7x-lighter-with-alpine-linux/
• http://blog.yohanliyanage.com/2015/05/docker-clean-up-after-yourself/

Credits

Inspired heavily by

• http://callistaenterprise.se/blogg/teknik/2015/04/10/building-microservices-with-spring-cloud-and-netflix-oss-part-1/