Devops Evolution

2016 FEDERAL FORUM

DevOps Evolution

Presented by Produced by

• Is DevOps a tooling or a cultural movement? • How does automation play a role? • How do you move beyond automation into continuous delivery? • Where should you get started?

SOURCE: HTTP://ROHITGHATOL.GITHUB.IO/ DEVOPS-GETTING-STARTED/#/1

A simple working definition: Infrastructure as code

DevOps tools are increasingly popular The Phoenix Project is the DevOps bible • Led by open source tools • A business novel modeled after • Most commonly provisioning tools Eliyahu Goldratt’s The Goal

Strong Culture + No Tools = Fail

ITIL Culture + DevOps Tools = Fail

…but people change is the long pole

When something goes wrong, is your instinct to:

Ctrl-Z: Back out the change Roll forward: Identify the and try again at the next OR problem, and quickly make maintenance window? the next change?

Designing for Correctness Designing for Resilience

Avoid failure at all costs Failure is a certainty • Focus on qualification and integration • During failure, service should be resilient • Create process gates to catch errors • Simulate failures to test • When a change doesn’t work, regroup • When a change doesn’t work, and use the process roll forward

• DevOps often gets lumped in with automation initiatives • DevOps is infrastructure as code • Automation is about executing workflows automatically

Both are important, but they are different. Be clear on what your objectives are.

• Most network automation discussions start with the network • You automate the act of doing something – The network is noun not a verb • A workflow is a set of tasks strung together to achieve some objective – Typical examples: provisioning and troubleshooting

• Deploy a server • Validate • Add an app instance • Deploy a service

Out of Box Troubleshooting Key Tools • Initial provisioning • Collecting diagnostics • Most tools represent an entry point to workflows – Ex: Ansible playbooks – Ex: Counters, stats – Ex: Logging tools lead • Planning • Remediation to troubleshooting – Ex: Inventory collection – Ex: Policy application – Ex: Server and app provisioning tools lead to edge policy

Data Distribution Normalization Logic • How do the elements • What language do the • If this, then that talk to each other? elements speak? • Requires rich set of • Not all elements • Even similar constructs triggers (or sensors) communicate can be formatted in different ways • Template-based set • Not all communication of actions is all the time

Rules Workflows

Sensors Audit Actions

Security

Networking

Event: “low disk on web301” StackStorm Unknown problem, need a human

Monitoring Resolve known Incident cases, fast. Is it Management /var/log? Clean up!

Wake up, buddy. Something real Web301 is is going on… “low disk” Service Engineer

Software Infrastructure

Source Object Executable Code File File File A C if a

B D 10111101 Library 11100001 00000011 01000100 Files 10011101 11111100

A1 B1 C1 D1 = Network1

A1 B2 C1 D1 = Network2

Modify Code Validate • Configuration management or • Network and SLA monitoring provisioning • Data collection and analysis Examples: CFEngine, Puppet, Chef, Ansible, Salt Examples: Nagios, OpenNMS, Zabbix

Unit Test Compile • Requires a test description language • Configuration rollout and deployment • Automated test harness Examples: Otto, Jenkins, Maven, Ant • Commercially unsolved for networking

Source Code Management • Configuration versioning • Aligned to management domains Examples: Git, Mercurial, Perforce

• Be clear about whether your company really wants DevOps • Fight the temptation to start with tools—people first! • Automation is about workflows, not infrastructure • Consider using your favorite tools to identify workflows • Build modularly and expand • Involve more than just the networking team