Microservices for Java Architects (Chicago, April 21, 2015)
-
Upload
derek-ashmore -
Category
Technology
-
view
481 -
download
1
Transcript of Microservices for Java Architects (Chicago, April 21, 2015)
Microservices for Java Architects
Given by Derek C. Ashmore
April 21, 2015
©2015 Derek C. Ashmore, All Rights Reserved 1
Who am I?
• Professional Geek since 1987
• Java/J2EE/Java EE since 1999
• Roles include: • Developer
• Architect
• Project Manager
• DBA
• System Admin
©2015 Derek C. Ashmore, All Rights Reserved 2
Discussion Resources
• This slide deck – http://www.slideshare.net/derekashmore
• Sample code on my Github – https://github.com/Derek-Ashmore/
• Sample Java Microservice (Moneta) – https://github.com/Derek-Ashmore/moneta
• Slide deck has hyper-links!
– Don’t bother writing down URLs
©2015 Derek C. Ashmore, All Rights Reserved 3
Agenda
The “What” and “Why” of microservices
Design Considerations
and Patterns
Packaging Options
Cross-cutting concerns
When to use microservices
Summary / Q&A
©2015 Derek C. Ashmore, All Rights Reserved 4
What are Microservices?
• No concrete definition
• Common microservice traits
– Single functional purpose
• Most/all changes only impact one service
• Not dependent on execution context – “loosely coupled”
– Independent process/jvm
– Standard Interface (typically Web Service/REST)
– Analogy: Stereo system, Linux utilities
©2015 Derek C. Ashmore, All Rights Reserved 5
Traditional Java EE Application Architecture
• Like a layer cake
• Highly cohesive
• Defined dependencies
• Deployed as one unit (war/ear)
• Limited Scalability
• Code Size
©2015 Derek C. Ashmore, All Rights Reserved 6
What is a Monolith?
• Hard to change – QA test cycles are long
– Change causes unintended consequences
• Hard to onboard new developers
• Married to your technical stack
• Harder to diagnose bottlenecks and memory issues
©2015 Derek C. Ashmore, All Rights Reserved 7
Microservices organize by domain
• Silo by business competency
• Divide and Conquer
• How to divide is the hard part
©2015 Derek C. Ashmore, All Rights Reserved 8
Microservices follow the Org Chart
• Conway’s Law
– Systems copy the structure of the organizations that build them
• Why?
– Corps need one manager in charge
– Reduces inter-group conflict
©2015 Derek C. Ashmore, All Rights Reserved 9
Refactoring into Microservices
• Large benefits to unified user interface
• Databases introduce unwanted coupling between services
©2015 Derek C. Ashmore, All Rights Reserved 10
Refactoring further
• Databases physically separated
• What to do with common data needs?
• Service call or
• Data copy
©2015 Derek C. Ashmore, All Rights Reserved 11
Alternate Database Strategy
• Centralized databases are hurdles for MS
– i.e. multiple apps use the same db
• MS-Central DB compromise
– Data items only maintained by *one* service
• All other services/apps consider that data item read-only
• Helps data corruption investigations
©2015 Derek C. Ashmore, All Rights Reserved 12
No Lock-in • Platform agnostic
• Fewer dependency conflicts
• Still have cross-cutting concerns
• “Toll” for first app
• Still have support concerns
• Others need to be able to support your work
13 ©2015 Derek C. Ashmore, All Rights Reserved
Easier Management / Higher Throughput
• Easier to manage large numbers of developers – Concentrate on
intelligently drawing service boundaries
– Manage/enforce service contracts
• Each service team works independently
• Team independence leads to higher development throughput
©2015 Derek C. Ashmore, All Rights Reserved 14
Isn’t this SOA?
• Yes – More or less
• No concrete definitions
• SOA == dumb endpoints and smart routes – ESB routes using Mule, Camel, etc.
– Transformations en route
• MS == dumb routes and smart end-points – Simple routes
• Usually REST or Soap calls via http(s)
• Persistent queue route at it’s most complex
• Analogy: Electrical supply for Stereo
©2015 Derek C. Ashmore, All Rights Reserved 15
Agenda
The “What” and “Why” of microservices
Design Considerations
and Patterns
Packaging Options
Cross-cutting concerns
When to use microservices
Summary / Q&A
©2015 Derek C. Ashmore, All Rights Reserved 16
Design considerations
• Service Boundaries (gerrymandering)
• Service call Failure / Unavailability
• Data Integrity
• Performance
©2015 Derek C. Ashmore, All Rights Reserved 17
Service Boundaries
• Core Services – Services responsible for maintaining a specific business area data – Usually named after Nouns
• Service is a system of record for a <blank> – Student, Course, Classroom, etc.
• Process Services – Services responsible for performing single complex tasks – Usually represents an Action or Process
• Service is responsible for processing <blank> – Student applications, Debt collection, etc.
– These services rely on core services
• Partitioning is an art – Too few same drawbacks as traditional architecture – Too many excessive network hops
©2015 Derek C. Ashmore, All Rights Reserved 18
Boundary Sanity Check
• Verbalize a mission statement in one sentence in business terms
– Examples
• This service is the system of record for Student information
• This service registers students for classes
• This service suspends students
• This service records student payments
• This service produces official transcripts
©2015 Derek C. Ashmore, All Rights Reserved 19
Microservices are not about size
©2015 Derek C. Ashmore, All Rights Reserved 20
….. Microservices are about having a single business purpose!
Designing for Failure
• Dependent services could be down
– Minimize human intervention
– Fail sooner rather than later
– Horizontal scaling / clustering is your first line of defense
– Coding patterns can help as a backup
• Common Patterns:
– Retry
– Circuit Breaker
– Dispatch via Messaging
©2015 Derek C. Ashmore, All Rights Reserved 21
Retry Pattern
©2015 Derek C. Ashmore, All Rights Reserved 22
• Best for asynchronous tasks • Limit the number of tries • Use sleep interval between tries • Only addresses temporary outages • Sample Retry Pattern implementation here. • Tooling Support:
– Apache CXF supports Retry – Spring Batch RetryTemplate – Apache HttpClient (Example here)
Circuit Breaker
©2015 Derek C. Ashmore, All Rights Reserved 23
Circuit Breaker (continued)
• Objective: Error out sooner – Conserves resources – Automatically “recovers” after a time period
• Modeled after home circuit • Works on thresholds
– Number of errors required to trip circuit – Amount of time required to attempt retry
• No formal tooling support for this pattern • Best embedded in interface clients / delegates • More information here. • Sample Circuit implementation here.
©2015 Derek C. Ashmore, All Rights Reserved 24
Dispatch via Messaging
©2015 Derek C. Ashmore, All Rights Reserved 25
• Place work instruction on persistent queue • If receivers are down, work stacks in queue • Work throttled by number of receivers • Queue can be JMS or AMQP • Tooling Support:
– JMS Api (easy API – many use natively) – Spring JMSTemplate or RabbitTemplate (producer)
Designing for Performance
• More network traffic
– Make services course-grained
– User Interfaces need a general manager
– Horizontal or Vertical Scaling helps
• Common Patterns:
– Back-ends for Front-ends (a.k.a. API Gateway)
– Dispatch via Messaging
– Expiring Cache
©2015 Derek C. Ashmore, All Rights Reserved 26
Back-ends for Front-ends
©2015 Derek C. Ashmore, All Rights Reserved 27
Back-ends for Front-ends (continued)
• Consolidates service calls for the browser
– Enhances performance
• Open web often not as performant as local LAN
• Also known as “API Gateway”
• Implications
– Don’t expose microservices directly to the browser
©2015 Derek C. Ashmore, All Rights Reserved 28
Expiring Cache
©2015 Derek C. Ashmore, All Rights Reserved 29
Expiring Cache (continued)
• Look up data once and cache it – Evict data from the cache after a defined time period – Sometimes known as “Cache Aside” – Reduces network calls for data – Trades memory for speed – More information here
• When to use – Only use with static data – Different clustered nodes “could” have different data for a short
time
• Tooling Support: – I recommend Google Guava – EHCache, Gemfire, and other tools available
©2015 Derek C. Ashmore, All Rights Reserved 30
Designing for Integrity
• Services are context independent
– Have no knowledge of how/when they are executed
• One service == One Transaction
– Two-phase commits/rollbacks are a much larger problem
• Common Patterns:
– Custom Rollback • Write your own reversing transaction
©2015 Derek C. Ashmore, All Rights Reserved 31
Context Independence
• No context assumptions
– Think in terms of a stereo system
• Warning Signs
– Spending time analyzing service call flow
• Your services likely make context assumptions
– Agonizing over which service should do a given activity
• Maybe you need a new service
©2015 Derek C. Ashmore, All Rights Reserved 32
Custom Rollback
©2015 Derek C. Ashmore, All Rights Reserved 33
Custom Rollback (continued)
• Reverses a transaction previously posted
• Only use this for multi-service transactions
– Keeping the transaction within one service is preferred
• This pattern is completely custom
– No special product support available
• More information here
©2015 Derek C. Ashmore, All Rights Reserved 34
Special Problems
• Versioning – Issues same as more traditional applications
– “Breaking” vs. “Non-Breaking” changes • Feature “additions” usually are “Non-Breaking”
• Trend away from breaking changes
– Versioning is a last result • It greatly increases number of contract tests
• Backward compatibility not always obvious to developers
• Prefer header versioning to URL versioning
• Explicitly error out for versions no longer supported
• More info here
©2015 Derek C. Ashmore, All Rights Reserved 35
Special Problems (continued)
• External Calls to 3rd Party Products
– Wrap external interfaces with a separate service
– Product upgrades or changes should only affect the interface service
– Log all requests/responses
©2015 Derek C. Ashmore, All Rights Reserved 36
Common code between services?
• Yes, but….
– Version it; services make decision as to when to upgrade
– Changes to common code can’t require the deployment of multiple services
• That ‘common code’ needs to be its own separate service
• Tends *not* to have business logic as that can change and impact multiple services
©2015 Derek C. Ashmore, All Rights Reserved 37
Agenda
The “What” and “Why” of microservices
Design Considerations
and Patterns
Packaging Options
Cross-cutting concerns
When to use microservices
Summary / Q&A
©2015 Derek C. Ashmore, All Rights Reserved 38
Packaging Support
• Microservices are deployed as a process
– For Java, embedded containers are easy
– Spring Boot
– Dropwizard
• Docker – standardizes the process deployment and environment
©2015 Derek C. Ashmore, All Rights Reserved 39
Spring Boot
• Packages Java EE application into *one* deployment jar – java –jar myApp.jar
• Support for health checks and other admin add ons via ‘Actuator’ add-on
• Supports either Jetty or Tomcat • Best for ‘Spring-mvc’ apps
– For non-spring apps, it’s swimming upstream
• Required artifacts – Maven
• spring-boot • spring-boot-starter-jetty (tomcat is available)
• spring-boot-starter-actuator (optional – health checks, etc.)
– Application class with public static void main() • Configuration coded (usually into the application class) • Will read application.properties (app-specific properties)
©2015 Derek C. Ashmore, All Rights Reserved 40
Spring Boot Actuator
• Actuator automatically adds endpoints
– Health (needs configuration)
– Metrics
– Mappings
– Trace
– Env
©2015 Derek C. Ashmore, All Rights Reserved 41
Dropwizard
• Packages Java EE application into *one* deployment jar – java –jar myApp.jar server myConfig.yaml
• Provides file configuration options (yaml format) – Database connection pools – Logging config – Port and other container specs
• Provides easy metrics/healthcheck support • Uses Jetty • Required artifacts
– Application class (with main()) – Maven: dropwizard-core, maven-shade-plugin
©2015 Derek C. Ashmore, All Rights Reserved 42
Docker • Is a “mini VM”
• runs a linux kernal
• Compare to shipping container
• Standard “connections” to outside world
• Supported formally by Oracle, Tomcat, Jboss, and many more
43 ©2015 Derek C. Ashmore, All Rights Reserved
Package Once, Run Anywhere!
Why Docker?
• Docker is Win-Win – Easier for OPS and system administrators
• All software looks the same
• Standard interface for disk and network resources – Containers can be “linked”
• Inherently automated
– Easier for developers • Fewer environment difference issues
• Less to communicate to OPS / system administrators
• Easy to leverage work of others (docker-hub)
– If you haven’t tried Docker yet – you should!
©2015 Derek C. Ashmore, All Rights Reserved 44
Sample Java/EE Microservice
• Moneta – Greek goddess of ‘memory’ – Open source: https://github.com/Derek-Ashmore/moneta
• Objective: – Provide a RESTful Web Service interface to a relational database
• Feature set: – Provides generic ‘core’ services – Returns Json-formatted data – Supports startRow and maxRows query options – Supports a security call-out – Built-in Dropwizard, Spring Boot, and War-file deployments
• Sample contract spec – currently read-only (writes in progress)
– /moneta/topics – lists ‘topics’ of information • E.g. – Topic Customer configured
– /moneta/topic/customers?startRow=5&maxRows=25 – /moneta/topic/customer/111-222-333
©2015 Derek C. Ashmore, All Rights Reserved 45
Agenda
The “What” and “Why” of microservices
Design Considerations
and Patterns
Packaging Options
Cross-cutting concerns
When to use microservices
Summary / Q&A
©2015 Derek C. Ashmore, All Rights Reserved 46
Cross-cutting Concerns
• Transaction tracking
• Security
• Contract Testing
• Same as traditional applications
– Health checks
– Logging consolidation
– Performance measurement
©2015 Derek C. Ashmore, All Rights Reserved 47
Correlation IDs • Provides context for
service calls or user actions
• Track using HTTP Header
• Log it on all messages / error reports
• Include it on all service calls or message dispatches
• Code sample here
• Spring Boot support has been requested
48 ©2015 Derek C. Ashmore, All Rights Reserved
Security
©2015 Derek C. Ashmore, All Rights Reserved 49
Security (continued)
• Keep User-level security to the UI
• Microservice security in layers
– Layer 1 – Network routing enforcement
• Limit access only to within the firewall
• Limit access to specific hosts or subnets
– Layer 2 – Use Service Accounts
• Similar to database access
©2015 Derek C. Ashmore, All Rights Reserved 50
Contract Testing
• Critical for MS architectures – Contract changes can break other services
– Bulkhead for rogue developers
– Makes individual services more disposable
• Consumer-based testing
• Tooling support – Apache HttpClient
– SoapUI
– ActiveMQ for JMS (embedded broker)
©2015 Derek C. Ashmore, All Rights Reserved 51
Agenda
The “What” and “Why” of microservices
Design Considerations
and Patterns
Packaging Options
Cross-cutting concerns
When to use microservices
Summary / Q&A
©2015 Derek C. Ashmore, All Rights Reserved 52
When to consider MS
• Starting out with MS isn’t recommended unless – Parts of the application will have extremely high volume
• Need to scale a portion of the application differently
• Note: MS isn’t all or nothing!
• Warning signs for app that’s too large – Unintended consequences after release
– High technical debt / design rot
– Release testing cycles abnormally large
– Need to coordinate large numbers of developers for a single code base • Large number == takes more than two pizzas to feed
©2015 Derek C. Ashmore, All Rights Reserved 53
Where’s the marketing fluff?
• Easier to manage – Maybe
• You *must* be good at contract management
• You *must* be good at specifying precisely what a microservice needs to do
• You *must* ensure that services make no assumptions on how they get invoked
• Easier for developers to “understand” applications – No – sorry
• It is easier to understand a particular ‘cog’ in the machine (e.g. the function of one service
• It is *not* easier to understand how microservices fit together to provide a particular piece of business functionality
©2015 Derek C. Ashmore, All Rights Reserved 54
Where’s the marketing fluff? (continued)
• Increased Development Throughput – Maybe
• Harder for business to ‘test’ a business function for a specific combination of microservices
• Developers work on *one* service at a time. • You *must* be good at error detection (unintended consequences) • The more assumptions a service makes about its execution context, the more
unintended consequences (e.g. errors) you are likely to have on deployment
• Services become disposable and can be “replaced” instead of “maintained / fixed”. – Maybe
• It’s more easily replaced than when using traditional architectures • Requires rigorous contract testing
– Can’t have the “replacement” act any differently than the original (except for the bug being fixed, of course)
• Requires architecture support for cross-cutting concerns – Can’t take a lot of time to implement / test
©2015 Derek C. Ashmore, All Rights Reserved 55
Common Mistakes
• Inappropriate Service Boundries – Services that are not truly loosely coupled
• One change Multiple services deployed
– Services that make ‘assumptions’ about execution context • Deployments cause unintended consequences
• Exposing MS to the browser – Causes performance issues – Use Back-ends for Front-ends – Keep MS traffic behind the firewall and on a fast
network
©2015 Derek C. Ashmore, All Rights Reserved 56
Why Now?
• SOA an old idea • Virtualization and Cloud advances (e.g. Docker, AWS,
etc.) – Less dependent on server set-up
• Hardware cheaper than it’s ever been – We’re not as concerned with the extra process overhead
for MS
• Infrastructure Automation advances – Additional tooling support(e.g. Chef or Puppet) – Easier to manage larger numbers of deployments
• Can you think of other possible reasons?
©2015 Derek C. Ashmore, All Rights Reserved 57
Future Directions
• What about Jigsaw?
– Java platform support for modularity
– Coming with Java9 (rumors are 2016)
• Early release is available
– Compared to Microservices
• More limited scaling flexibility
• More limited technical stack
• Chief benefit over MS is reduction in process memory – Less important as memory is cheaper these days
©2015 Derek C. Ashmore, All Rights Reserved 58
Future Directions
• What about OSGi? – OSGi supplies modularity to JVM
– Compared to Microservices • More limited scaling flexibility
• More limited technical stack
• Chief benefit over MS is reduction in process memory – Less important as memory is cheaper these days
– Low adoption to date • Jobs oscillate around 0.01% (indeed.com)
• Google Trends shows steady decline
©2015 Derek C. Ashmore, All Rights Reserved 59
Future Directions (continued)
• Trend toward product releases designed to be deployed as microservices
– Already seeing increase on GitHub
– Takes PAAS to a new level
• Will Microservices become main stream?
– Too early to tell
©2015 Derek C. Ashmore, All Rights Reserved 60
Further Reading
• Microservices reading list – http://www.mattstine.com/microservices
• Microsoft’s Cloud Design Patterns – https://msdn.microsoft.com/en-us/library/dn600223.aspx
• Moneta Java microservice example – https://github.com/Derek-Ashmore/moneta
• This slide deck – http://www.slideshare.net/derekashmore
©2015 Derek C. Ashmore, All Rights Reserved 61
Questions?
• Derek Ashmore: – Blog: www.derekashmore.com
– LinkedIn: www.linkedin.com/in/derekashmore
– Twitter: https://twitter.com/Derek_Ashmore
– GitHub: https://github.com/Derek-Ashmore
– Book: http://dvtpress.com/
©2015 Derek C. Ashmore, All Rights Reserved 62
May 14-15, 2015 http://www.chicagocoderconference.com/