DATA DRIVEN DEVOPS™
“Statistics and Data get
everyone in the game”
© 2015 Brian McCallion. All rights reserved.
http://bronzedrum.com
646 308-1257
attribution: “A time based / event series interactive visualization using d3.js” http://marmelab.com/EventDrops/
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
INTRODUCTION
Brian McCallion, Founder Bronze Drum Consulting
Name, Role
Your Questions / Expectations Around DevOps
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
Six Principles of Data Driven DevOps
1. Begin with a simple set of automated real-time metrics
2. Zero manual reporting preferred over too many metrics
3. Quality of software is measured continuously as a function of: cost,
coupling, fault tolerance, elasticity, time to repair / time to delivery
4. Lessons learned = new metrics
5. Bottoms up automated reporting
6. System logs, Software logs, Cloud API calls, external task and bug
tracking systems are all sources of metrics
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
DEVOPS AS A SHARED STATISTICAL FEEDBACK
LOOP
John Boyd Observe,
Orient, Decision, Action
(OODA Loop)
Decisions are made based
on the perceived utility
value of taking an action
Continuous Optimization v
SLAs
attribution: Patrick Edwin Moran
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
Data
CloudTrail
CloudWatch
CloudWatch Logs
Agile Collaboration
Tools
(Atlassian Jira,
Jenkins, Github,
Slack, CodeDeploy,
CodePipeline
Visualization
Sources
attribution: “A time based / event series interactive visualization using d3.js” http://marmelab.com/EventDrops/
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
HOW DO WE MODEL DEVOPS?
Monitoring becomes a primary, cross-cutting concern
Real-time KPIs from Development Lifecycle, Systems,
Application, and Cloud API Data
Ingest, filter, count and monitor real-time data from
diverse sources
Examples
Real time cost data
Performance comparison
Web Server logs 404 errors, successful
transactions, service unavailable
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
Why Model DevOps Data?
Always know exactly what’s going on with your project, your systems, your applications
Control? Are we losing control?
→ We have more control than ever before, plus the power of self-directed action
→ Architecture reflects Organizational Structure, Data reflects process
When we all have the data, we all know Who’s On First? What Inning, and we play
as a team.
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
WHY DATA FIRST?
Without Data we don’t know what is going on
-Are we making progress?
-How many manual v automated deployments today?
Think of Continuous Integration as OODA Loop
Decision to Action → Accelerate
SLA → Continuous Optimization
Illusion of Control → “the beautiful game”
OODA Gets in the Head of Our Opponent --> Disrupt Status Quo
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
Patterns and Anti-Patterns of Delivery and Deployment
Anti-pattern Pattern
Manual intervention. “Wait until
[name] gets back from lunch….”
Automated process
Hiring DevOps experts, thinking of
DevOps as outside our
organization*
*Embedding experts. Expanding
roles and responsibilities and
practices around well-understood
outcomes.
Taking extended time to discover
simple problems
Run the build quickly to flush out
issues continuously
One off-changes Small continuous improvements to
build process
Tight coupling of application
components
Decouple application components.
Microservices each use a separate
data store.
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
OODA IN ACTION
DHS (and yes, Netflix) Runs Chaos Monkey in Production
→ Accelerates optimization of processes by turning the wheel faster
Blue Green Deployment Model
→ Continuously test infrastructure as code v Production
→ Each deployment tests if our error and other metrics are the right KPIs
→ Our understanding improves; our system gets smarter, tighter, faster,
better
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
THE BLUE GREEN GOLD STANDARD
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
AWS
CLOUDFORMATION + User Data
In Less thirty minutes Create an entire
VPC, with Autoscaling Alfresco cluster,
RDS Multi-zone, Apache Web Servers,
TomCat !
Link to the Alfresco CloudFormation
template.
Note how CloudFormation and User
Data are used in tandem.
Use Case
Challenge
● Autoscaling
Clusters
● Zero Database
License Cost
● Encrypted
Datastore
Alfresco + AWS RDS Aurora Achieves Billion Document Milestone
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
CLOUD AND DEVOPS: STREAMS OF
STATISTICS
Web Services What? Source Super Power
Cloudtrail Security and Cloud
Capability
Benchmarking
All AWS API calls Omniscience
VPC FlowLogs Network events at
VPC Leve
VPC Network Events
Streamed to
CloudWatch Logs
Long awaited
visibility into VPC
network
Cloudwatch Logs Stream logs into AWS
from any system
Any logs from
Datacenter, Cloud
Observe / Orient
CloudWatch Repository for
statistics
Any system Observe / Orient
CloudWatch Alarms Message, HTTP,
Shutdown, Terminate
Cloudwatch
threshold reached
Decision / Action
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
What’s New in VPC?
VPC Peering
Enables low latency, virtually unlimited bandwidth between VPCs
VPCs can be in different accounts (in the same region)
Simple workflow and routing between subnets in different VPCsIn many
cases we seek to migrate or move workloads across accounts or VPCs
→ This used to be difficult
→ Now it’s easy, zero cost, no internet gateway or VPN required!!
© 2015 Brian McCallion. All Rights Reserved.http://bronzedrum.com
Scaling Alfresco on AWS
Another way to do
“DevOps” is to offload
the management, or
“undifferentiated
heavy lifting” to a
service like Amazon’s
Relational Database
Service
LET’S GET STARTED What’s next? How can we
help?

Data Driven DevOps: from Culture to Gamification

  • 1.
    DATA DRIVEN DEVOPS™ “Statisticsand Data get everyone in the game” © 2015 Brian McCallion. All rights reserved. http://bronzedrum.com 646 308-1257 attribution: “A time based / event series interactive visualization using d3.js” http://marmelab.com/EventDrops/
  • 2.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com INTRODUCTION Brian McCallion, Founder Bronze Drum Consulting Name, Role Your Questions / Expectations Around DevOps
  • 3.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com Six Principles of Data Driven DevOps 1. Begin with a simple set of automated real-time metrics 2. Zero manual reporting preferred over too many metrics 3. Quality of software is measured continuously as a function of: cost, coupling, fault tolerance, elasticity, time to repair / time to delivery 4. Lessons learned = new metrics 5. Bottoms up automated reporting 6. System logs, Software logs, Cloud API calls, external task and bug tracking systems are all sources of metrics
  • 4.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com DEVOPS AS A SHARED STATISTICAL FEEDBACK LOOP John Boyd Observe, Orient, Decision, Action (OODA Loop) Decisions are made based on the perceived utility value of taking an action Continuous Optimization v SLAs attribution: Patrick Edwin Moran
  • 5.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com Data CloudTrail CloudWatch CloudWatch Logs Agile Collaboration Tools (Atlassian Jira, Jenkins, Github, Slack, CodeDeploy, CodePipeline Visualization Sources attribution: “A time based / event series interactive visualization using d3.js” http://marmelab.com/EventDrops/
  • 6.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com HOW DO WE MODEL DEVOPS? Monitoring becomes a primary, cross-cutting concern Real-time KPIs from Development Lifecycle, Systems, Application, and Cloud API Data Ingest, filter, count and monitor real-time data from diverse sources Examples Real time cost data Performance comparison Web Server logs 404 errors, successful transactions, service unavailable
  • 7.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com Why Model DevOps Data? Always know exactly what’s going on with your project, your systems, your applications Control? Are we losing control? → We have more control than ever before, plus the power of self-directed action → Architecture reflects Organizational Structure, Data reflects process When we all have the data, we all know Who’s On First? What Inning, and we play as a team.
  • 8.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com WHY DATA FIRST? Without Data we don’t know what is going on -Are we making progress? -How many manual v automated deployments today? Think of Continuous Integration as OODA Loop Decision to Action → Accelerate SLA → Continuous Optimization Illusion of Control → “the beautiful game” OODA Gets in the Head of Our Opponent --> Disrupt Status Quo
  • 9.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com Patterns and Anti-Patterns of Delivery and Deployment Anti-pattern Pattern Manual intervention. “Wait until [name] gets back from lunch….” Automated process Hiring DevOps experts, thinking of DevOps as outside our organization* *Embedding experts. Expanding roles and responsibilities and practices around well-understood outcomes. Taking extended time to discover simple problems Run the build quickly to flush out issues continuously One off-changes Small continuous improvements to build process Tight coupling of application components Decouple application components. Microservices each use a separate data store.
  • 10.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com OODA IN ACTION DHS (and yes, Netflix) Runs Chaos Monkey in Production → Accelerates optimization of processes by turning the wheel faster Blue Green Deployment Model → Continuously test infrastructure as code v Production → Each deployment tests if our error and other metrics are the right KPIs → Our understanding improves; our system gets smarter, tighter, faster, better
  • 11.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com THE BLUE GREEN GOLD STANDARD
  • 12.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com AWS CLOUDFORMATION + User Data In Less thirty minutes Create an entire VPC, with Autoscaling Alfresco cluster, RDS Multi-zone, Apache Web Servers, TomCat ! Link to the Alfresco CloudFormation template. Note how CloudFormation and User Data are used in tandem. Use Case Challenge ● Autoscaling Clusters ● Zero Database License Cost ● Encrypted Datastore Alfresco + AWS RDS Aurora Achieves Billion Document Milestone
  • 13.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com CLOUD AND DEVOPS: STREAMS OF STATISTICS Web Services What? Source Super Power Cloudtrail Security and Cloud Capability Benchmarking All AWS API calls Omniscience VPC FlowLogs Network events at VPC Leve VPC Network Events Streamed to CloudWatch Logs Long awaited visibility into VPC network Cloudwatch Logs Stream logs into AWS from any system Any logs from Datacenter, Cloud Observe / Orient CloudWatch Repository for statistics Any system Observe / Orient CloudWatch Alarms Message, HTTP, Shutdown, Terminate Cloudwatch threshold reached Decision / Action
  • 14.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com What’s New in VPC? VPC Peering Enables low latency, virtually unlimited bandwidth between VPCs VPCs can be in different accounts (in the same region) Simple workflow and routing between subnets in different VPCsIn many cases we seek to migrate or move workloads across accounts or VPCs → This used to be difficult → Now it’s easy, zero cost, no internet gateway or VPN required!!
  • 15.
    © 2015 BrianMcCallion. All Rights Reserved.http://bronzedrum.com Scaling Alfresco on AWS Another way to do “DevOps” is to offload the management, or “undifferentiated heavy lifting” to a service like Amazon’s Relational Database Service
  • 16.
    LET’S GET STARTEDWhat’s next? How can we help?