project sherpa: how rightscale went all in on docker

12
PROJECT SHERPA: HOW RIGHTSCALE WENT ALL IN ON DOCKER

Upload: rightscale

Post on 15-Apr-2017

217 views

Category:

Technology


1 download

TRANSCRIPT

PROJECT SHERPA:

HOW RIGHTSCALE WENT ALL IN

ON DOCKER

• Tim Miller, VP Engineering

• Ryan O’Leary, Senior Director of Product

• Mark Dotson, Principal Systems Administrator & Lead Sherpa

• Ryan Williamson, Senior Manager, Engineering

• Tony Spataro, Senior Systems Architect

Panelists

Docker Use is Exploding

It’s Not Just Dev

• Goals, Timeline, Process

• Results

• Demo

• What Went Well/Hard

• Q & A Panel

Agenda

• RightScale Product

Strategy

• RightScale Bank

Account

• Productivity Increase

• Technical Staff’s

Career

Project Sherpa is good for everyone!

• 1028 running instances

• 670 running “dynamic” apps would be containerized

• Dynamic apps are RightScale developed apps, potentially new versions

released weekly, 52 in total.

• 358 running “static” apps would not be migrated (no ROI)

• SQL databases, Cassandra rings, MongoDB clusters, Redis,

Memcached, RabbitMQ systems, Syslogs, HTTP load balancers,

Collectd aggregators, proxy servers, and VPN endpoints.

Base Camp

• Expecting a 5 week hike, but

experienced a 7 week one

• 90% of the Eng Team for 5 weeks,

10% weeks 6,7

• 100% of OPS for 6 weeks, 50%

week 7

• Paused feature releases during

that time

• Split services into 3 groups

• Implemented “Bay of

Containers”

The Climb

• Converted 49 of 52 services to containers

• Reduced instance count by 55%

• Translated to a weighted avg savings of 53%

• 44% savings in Production

• 63% savings in Staging

• 74% savings in Integration

• Faster Developer Workflow

• Devs & QA now using laptops for dev/test vs. cloud based env

• Seeing velocity increases due to this less complex, self-contained

setup

• Reworking all our automation tools

The Summit

• Time while partially containerized in staging/prod was painful

• OPS/Production Support had to have 3 run books

• Developers were blocked

• Hotfixes to production took more time (2-3x)

• Minimize time you spend in this transition.

• If you are touching legacy apps, be ready for a bunch of Tech Debt

requests. Initially took the stance of “No”, but did end up letting some go

through.

• Don’t underestimate the effort around reworking dev/test/release tools.

• Needed more training for devs on the new branching and build workflows

• Experienced performance issues with Docker & Ubuntu 14.04 as Docker

Engine/host.

• Moving to CentOS 7 will evaluate later versions of Ubuntu

Challenges Encountered/Lessons Learned

• Contact us for a personalized demo

[email protected]

Q&A