33
How do you implement Continuous Delivery? Part 5: Deployment Patterns

How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Embed Size (px)

DESCRIPTION

In Part 5 of our series on putting CD into practice, we examine a few deployment patterns with their pros and ons

Citation preview

Page 1: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

How do you implement

Continuous Delivery?

Part 5: Deployment Patterns

Page 2: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Key Principle:

Low–risk releases are incremental

Page 3: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Why? Low–risk releases are incremental

§  Big-bang releases that involve multiple dependent components,

database changes and/or business logic changes are highly volatile.

§  Instead incremental releases, where the new functionality and all

dependent services are thoroughly tested, and rollbacks are easier,

are low-risk.

§  Let’s explore some low-risk incremental deployment patterns…

Page 4: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Blue-Green

Deployment Pattern

Page 5: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

http://martinfowler.com/bliki/BlueGreenDeployment.html

Blue-Green Deployment Pattern

§ Minimizing downtime, while doing the “cut-over” from testing to

release is one of the key challenges with automating deployment.

§  The blue-green deployment approach does this by ensuring you

have two identical production environments.

§  It also helps you to rapidly rollback in the event of a failure.

Page 6: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

Blue Environment

Release 1

Green Environment

Blue-Green Deployment Pattern

At any time only one production environment, let's say, blue, is live

Page 7: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

Blue Environment

Release 1

Green Environment

Release 2

Blue-Green Deployment Pattern

As you prepare a new release of your software you do your final stage of testing in the green environment.

Page 8: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

Blue Environment

Release 1

Green Environment

Release 2

Blue-Green Deployment Pattern

Once the software is working in the green environment, you switch the router so that all incoming requests go to the green environment

Page 9: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

Blue Environment

Release 3

Green Environment

Release 2

Blue-Green Deployment Pattern

The blue environment is now available for you to deploy your next release.

Page 10: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Phoenix

Deployment Pattern

Page 11: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

http://kief.com/configuration-drift.html http://martinfowler.com/bliki/PhoenixServer.html

Phoenix Deployment Pattern

§  Phoenix servers are those that you virtually tear down at regular

intervals.

§  Configuration drift describes inconsistencies between servers caused

by ad-hoc changes over time.

§  Phoenix servers are a great way to avoid configuration drift, as they

are rebuilt from a common template, and are not kept running for

long enough for much configuration drift to accumulate.

Page 12: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

Consider Release 1 on R1 Environment

R1 Environment

Release 1

Phoenix Deployment Pattern

Page 13: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

R1 Environment

Release 1

R2 Environment

Release 2

Ready Release 2 on the R2 Environment Phoenix Deployment Pattern

Page 14: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

R1 Environment

Release 1

R2 Environment

Release 2

Switch the router to the R2 Environment Phoenix Deployment Pattern

Page 15: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

Kill the R1 Environment

R2 Environment

Release 2

Phoenix Deployment Pattern

Page 16: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

http://martinfowler.com/bliki/BlueGreenDeployment.html

R2 Environment

Release 2

R3 Environment

Release 3

Continue the process with the R3 Environment Phoenix Deployment Pattern

Page 17: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Environment Promotion

Pattern

Page 18: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

?

Environment Promotion Pattern

§ With this pattern, a new environment is created for each

software release, and the environment itself is promoted

through the stages of the pipeline.

§  This ensures that the actual environment has been tested, rather

than only the changes to the configuration.

§  This pattern may be inappropriate when an environment needs

to be integrated with different external services at different

stages of the pipeline.

Page 19: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

?

The R2 environment created for Release 2 of the application, is tested in the QA stage

R1 Environment

Release 1

Production Router

UAT

R2 Environment

Release 2

QA Router

Environment Promotion Pattern

Page 20: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

?

The R2 environment is connected to the UAT router, and Release 2 goes through user acceptance testing.

Production Router

QA Router UAT

R1 Environment

Release 1

R2 Environment

Release 2

Environment Promotion Pattern

Page 21: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

?

Once the R2 environment and its software release have passed UAT, the production router is configured to send traffic to it, and the R1 environment is destroyed.

Production Router

QA Router Staging

R2 Environment

Release 2

Environment Promotion Pattern

Page 22: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Canary

Release Pattern

Page 23: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

http://www.informit.com/articles/article.aspx?p=1833567 http://techcrunch.com/2011/05/30/facebook-source-code/

Canary Release Pattern

§  This is a variation of blue-green deployment and is applicable

when running a cluster of servers.

§ With this pattern, rather than upgrading a whole cluster to the

latest version all at once, you do it incrementally.

§  This allows you to get feedback from a small subset of users prior

to a complete rollout

§  Like canaries in a coal mine, if a problem is discovered at the

initial stages, the build goes no further.

Page 24: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

Consider a cluster of servers Canary Release Pattern

R1 R1 R1 R1 R1 R1 R1 R1

R1 R1 R1 R1 R1 R1 R1 R1

R1 R1 R1 R1 R1 R1 R1 R1

Page 25: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

Canary Release Pattern

R1 R1 R1 R1 R1 R1 R2 R2

R1 R1 R1 R1 R1 R1 R1 R1

R1 R1 R1 R1 R1 R1 R1 R1

The build is first routed to a small section of servers/users

Page 26: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

Canary Release Pattern

R1 R1 R1 R1 R1 R1 R2 R2

R1 R1 R1 R1 R1 R1 R1 R1

R1 R1 R1 R1 R1 R1 R1 R1

The release is validated with performance testing and multi-variant testing

Page 27: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

Canary Release Pattern

R2 R2 R2 R2 R2 R2 R2 R2

R2 R2 R2 R2 R2 R2 R2 R2

R2 R2 R2 R2 R2 R2 R2 R2

Only after the release feedback is positive, is it rolled out to all servers/users

Page 28: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Dark Launching

Page 29: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

http://www.facebook.com/note.php?note_id=96390263919

Dark Launching §  This involves releasing a new feature to a subset of users, with

minimal UI changes, while exercising all the parts of your

infrastructure involved in serving that feature.

§  This pattern is useful for massive, large-scale deployments to

simulate load/stress testing.

§  Dark launching exposes pain points and areas of the

infrastructure that need attention prior to the actual launch.

Page 30: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

Rollout the release to all, with the new feature within it being released to only a subset of servers/users

Dark Launching

R1 Release R2 Release

New Feature

R2 Release

New Feature

Page 31: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Router

Only after satisfactory load/stress testing and feedback on the new feature, is the new feature rolled out to all servers/users

Dark Launching

R1 Release R2 Release

New Feature

R2 Release

New Feature

Page 32: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Data Management

Stay tuned for Part 6…

Page 33: How do you implement Continuous Delivery?: Part 5 - Deployment Patterns

Deploy a great product faster. Agile teams deliver working software early and often. Go automates and streamlines the build-test-release cycle for worry-free, continuous delivery of your product.

Learn More See how Go can help you in your CD journey

goContinuous Delivery