Speed up Continuous Delivery with BigData Analytics

#JenkinsWorld
Speed up Continuous Delivery
with BigData Analytics
Luca Milanesio

@gitenterprise @gerritreview #JenkinsWorld
• Founded in the UK
• HQ in London, Offices in the UK and Italy
• Committed to OpenSource
About GerritForge

• Co-founder and Director of
GerritForge
• over 20 years in Agile
Development and ALM
• Contributor to many OpenSource
projects
• Mind and hands of @gitenterprise
and @gerritreview Twitter feeds
About Luca Milanesio

NOT a lecture on Continuous Delivery
NOT a lecture on Code Review
NOT a lecture on BigData
What NOT to expect?

SO WHAT?
#!@$!

SHOW
a Continuous Delivery scenario

LEARN
together from its problems

EXPERIMENT
a new approach using BigData

DISCUSS
results together and iterate

Continuous Delivery Pipeline
Push Build
Unit Tests
Push Build
Unit Tests
Automated
Tests
Push Build
Unit Tests
Automated
Tests
Beta Users
Tests
User's feedback Publish to all Users
Failed
Failed

Code Review & CD Validation
A1
refs/for/master
master
C2 C2*
A1 C2 (local)
master
C2*
push
Verified: -1
(build failed)
push
Verified: +1
(build OK)
Reviewed: +2
CI Build: OK
C2*
CI Build: OK
amend

Git/Gerrit
(Code Review)
Components
Jenkins
(Continuous
Integration & Delivery)

DEMO
See the pipeline in action

We use micro-services
–100+ Components
–200+ Repositories
–300+ People and 100,000+ of review comments
–Cross-components [implicit] dependencies
–Lots of build chains
–10,000+ of builds
Now feel the pain :-(

Problem #1 : Tons of logs slows us down
"Build was so fast a week ago … why is it getting so
slow now?"
100Ks of logs are tough for Jenkins

"It was working until 5 mins ago … then someone
changed something"
 Rings any bell?
Problem #2 : Ripple effect

In order to deliver Feature ABC
I NEED
Ver. N of Service-A with
Ver. M of Service-B and
Ver. K of Service C
Problem #3 : Dependencies

Release small changes is GOOD
Release often every change is GOOD
BUT
How long it takes the "Big Picture" and End-to-End?
Where is my bottleneck?
Problem #4 : Cycle time

Problems #1 + #2 + #3 + #4 =
DevOps meltdown 
Problem #6 : all together

• Collect all review events
• Collect all logs
• Channel them to a central store
• Crunch and Crunch continuously
• Never delete
• Process, inspect and learn
BigData to the rescue

Continuous Delivery – Analytics Pipeline
Data
Aggregation
Jobs

Continuous Delivery – Analytics Dimensions
Projects Commits
People Reviews System Metrics

Continuous Delivery – Tools Dimensions
SCM / Code Review
CI / CD

DEMO
Continuous Delivery Analytics

• Collect more data (System, JVM, Review scores)
• Real-time statistics
• Events correlation
• Feedback loop to Code-Review
• Production data harvest
• More?
Future directions

Want to know more?
www.gerritforge.com/contact

• GerritForge's blog
www.gitenterprise.me
• Gerrit Code Review
www.gerritcodereview.com
• Jenkins
www.jenkins.io
• RabbitMQ
rabbitmq.com
Resources
• Apache Flume
flume.apache.org
• Hadoop HDFS
hadoop.apache.org
• Apache Spark
spark.apache.org
• Kibana
github.com/elastic/kibana

Speed up Continuous Delivery with BigData Analytics

More Related Content

Speed up Continuous Delivery with BigData Analytics