Microservices Tutorial Session at JavaOne 2016

Microservices
A Tutorial Session
JavaOne 2016

Recipe: Microservices
(makes 5-10 microservices)
Ingredients:
● A Monolithic API, Too Big To Handle
Instructions:
1. Find the seams in your application - by operation, business area, or team.
2. Split your API into multiple smaller API’s, each on its own server.
3. Figure out how to deploy them w/o breaking everything.
4. Figure out how they’ll talk to each other w/o breaking more things.
5. Figure out how complex data operations will work across multiple data stores.

1. Small & Maintainable
2. Private data
3. Generally Isolated
Microservice Properties

1. Small & Maintainable
2. Private data
3. Generally Isolated
Microservice Nice-To-Haves

1. Findable
2. Scalable
3. Resilient
4. Monitorable
Microservice Nice-To-Haves

Microservices That Speak To Each Other Via
Asynchronous Communication
Microservice Real Nice-To-Haves

Let’s See a
Real
Transaction

Web Client: PUT issues/123 Assignee=10
Issues Service: Read from issues db
Issues Service: GET users/10
Issues Service: POST users/notify/10
Issues Service: Write to issues db
Issues Service: Return response with result
Web Client: Read response & status then handle it

Web Client: PUT issues/123 Assignee=10
Issues Service: Read from issues db
// block for synchronous transaction
// block for synchronous transaction
Issues Service: Write to issues db
Issues Service: Return response with result
Web Client: Read response & status then handle it

Blocking Isn’t
Isolated, Scalable
Or Resilient

1. Easy!
2. Gratifying.
3. Testable.
Synchronous Messaging Pros

1. Prevents microservices from being isolated,
scalable, and resilient.
2. Hope you like waiting.
Synchronous Messaging Cons

Microservices
Should Talk
Asynchronously

1. Non-Blocking IO (eg RxJava, Netty)
2. Queues (eg SQS)
3. Event Streams (eg Kafka, Kinesis)
4. Actors (eg Akka)
Talking Asynchronously

1. Synchronous REST with Async Clients
2. Http 202
3. Websockets
4. Http/2
Responding Asynchronously

That’s What
Microservices Are All
About

Why
Might This Be A
Good Idea?

(Assuming You’re Planning
to Split Up your Monolith)

1. Finding code and refactoring it is pretty easy.
2. No network lag
3. No format conversion errors
4. Deploy in one go
Monoliths - Not So Bad

That’s All
Thanks For Listening!

1. Hard to scale up just one feature
2. Hard to deploy just one change
3. Not suitable for more than a small team
Monoliths - Not So Scalable Either

Organizations Are
Constrained To Produce
Designs That Match Their
Communication Structure

Don’t Split Monoliths
Across Your
Communication
Structures

(we kind of covered this already)
Microservices Pros

1. Network Lag
2. Multiple Data Formats
3. Service Resolution
4. Troubleshooting
5. More Deployment Work
6. More Monitoring
Microservices Challenges

No Problem.
Let’s Get Started On
How
You’ll Build Em

How Will You Get
Your App In
Production?

1. Self-Running Jar’s
2. VM Images
3. Docker Images
Deployment Formats

1. Use A CI To Build / Test / Merge in CI
2. Use A CD To Deploy (Push-Button / Auto)
3. VM Deployment (eg Spinnaker)
4. Docker Deployment (eg ECS, K8s, Marathon)
5. Cluster Management (eg K8s, Mesos, Swarm)
Deployment Strategies

Yes
Should I Deploy To The Cloud?

YES OH YES
Should I Test It First?

1. Service Resolution (K8s, Consul)
2. Scaling
3. Visibility
4. Authentication
5. Monitoring
Microservice Management

Let’s Talk
Microservices
Development

(why switch now?)
Choosing A Framework

1. Spring Boot
2. Lagom
3. Ratpack
4. Dropwizard
Choosing A Framework

1. Good Ol’ SQL (eg MySQL, Postgresql)
2. Document-type NoSQL
3. Fast Key-Value Caches
Choosing A Data Store

1. CRUD operations endpoints & tables
2. CRUD endpoints and immutable tables
3. Event-Sourced Data
Choosing A Data Scheme

Let’s Talk About
Immutable Tables

GET /issues
GET /issues/{id}
POST /issues
PUT /issues/{id}
Issue Endpoints

How Do These Events
Affect The Database?

2. PUT /issues/1 assignee=10
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:16 | Config ELB | 10 | false |
+------+-------------+------------+----------+-------+

3. PUT /issues/1 done=true
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:24 | Config ELB | NULL | true |
+------+-------------+------------+----------+-------+

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
Not Bad.

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
Do You Know How We Got Here?

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
Why is ‘assignee’ NULL?

Mutable Table Rows
Lose History

Immutable Table
Rows KeepTheir History

Let’s Try To
Lock Down
Our State

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:16 | Config ELB | 10 | false |
+------+-------------+------------+----------+-------+

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:16 | Config ELB | 10 | false |
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+

1. GET /issues/1
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:16 | Config ELB | 10 | false |
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+

How Do You Make An
Append-Only
Table?

One: Don’t Let Your DB
User Make
Changes

Grant select, insert on
issues to my-db-user;

create table issues (
id serial,
created timestamp default now(),
creator_id int,
issue_id int default nextval(‘iseq’),
title text,
assignee int,
done boolean default false
)

Let’s Talk About
Event-Sourcing

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:16 | Config ELB | 10 | false |
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:24 | Config ELB | 10 | true |
+------+-------------+------------+----------+-------+

+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:16 | Config ELB | 10 | false |
+------+-------------+------------+----------+-------+
| 1 | 09-18 18:24 | Config ELB | 10 | true |
+------+-------------+------------+----------+-------+
Events
States

Now We’re Storing
Events,
Not States

create table issue_events (
id serial,
created timestamp default now(),
issue_id int default nextval(‘iseq’),
originator text,
payload text
)

1. POST /issue/1/event ‘Originator: 4a48239-8a..’
payload=’<Update val=”done=true”>’
+----+-------------+----------+------------+---------+
| id | created | issue_id | originator | payload |
+----+-------------+----------+------------+---------+
| 14 | 09-18 18:50 | 1 | 4a482... | <...> |
+----+-------------+----------+------------+---------+

Create Events And
Simulate The State

1. Create-Issue
Issue(“Config ELB”, null, false);
Real Events
Virtual States

1. Create-Issue
2. Assign-Issue
Issue(“Config ELB”, 10, false);
Real Events
Virtual States

1. Create-Issue
2. Assign-Issue
3. Complete-Issue
Issue(“Config ELB”, 10, true);
Real Events
Virtual States

1. High Write Performance
2. Potential for Command/Query Separation
3. Auditable
4. Replayable
5. Undo-able
6. Monitorable
Reasons For Event-Sourcing

It’s Like Having Control
Over The Versions Of
Your State Changes

It’s Like Having Control
Over The Versions Of
Your Data

1. Frankly, It’s Weird
2. Requires Events. No Events, No Event-Sourcing.
3. As Of Sept 2016, It’s Still Non-Standard
Reasons Against Event-Sourcing

That About Sums Up
Microservice
Development

Okay, Actually That’s The
Entire Session
Unless There’s More Time

Web Applications for the REST of Us: An Introduction to
Ember.js, Akka, and Spray (Craig Tataryn & Sean Kowaski)
A Practical RxJava Example with Ratpack (Laurent Doguin)
One Microservice Is No Microservice: They Come In Systems
(Markus Eisele)
Make Sure To See...

Just Kidding, There’s
Another Section

Let’s Talk
Asynchronous
Architectures in AWS

Do You Want To Use
Caches To Prevent
Unnecessary
Blocking Calls?

Do You Want To Insert
Rows And Publish
Events In One Stroke?

Do You Want To
Mass-Publish One Event
To All Microservices?

Note About Your Solution:
I’m Not Saying It’s AWS

Okay, We Are Really
Done This Time

Microservices Tutorial Session at JavaOne 2016

More Related Content

Viewers also liked

Similar to Microservices Tutorial Session at JavaOne 2016

Recently uploaded

Microservices Tutorial Session at JavaOne 2016