2017-08-04#

Ran Across Today#

Which brings us to the second reason for this post, the difference between “Architecture” and “Design”. In a nutshell, architecture is a type of design where the focus is quality attributes and wide(er) scope whereas design focuses on functional requirements and more localized concerns.

Randy Shoup - Techniques for dealing with shared data, joins, and transactions in a microservices architecture#

https://www.infoq.com/presentations/microservices-data-centric

Scratch Notes...

Software combines #

Organizes
  • Small Teams directly defined within a particular function of the business.
  • Teams need other teams.

Processes - Test Driven Development#

  • Testing increases velocity
  • Tests make better code
  • Provides courage to try new thing, knowing you can refrator something and know it will work or not.
  • Tests allow fast failures

We do not have time to do it right? - Do you have time to do it more than once?

  • The more time constraints the more it is better to build it right (80/20 rule)
  • Better to build one thing right than two things half-right.

Continuous Delivery #

  • Release smaller sprints.
  • Allows rapid experiments

DevOps#

  • You build it, you Run it.
  • End-to-end ownership of what you write - Full lifecycle of software.
  • No separate QA team
  • No separate Deployment team

Evolution to microservices#

  • ebay 5th generation
  • Pearl
  • C++ 3.4 million lines of code
  • Java - Several different Java Apps
  • Polyglot of microservices.

Twiter similar.#

Amazon #

  • Mono perl and c++ OBIDOUS
  • Java Scala

No place started with Microservice and no place at this scale is NOT using Microservice.

Microservices may not be the right for startups. Monos are ok.

Microservices #

  • Scope is single purpose
  • Moduler and independenrt
  • SOA done properly (Bounded Context)

Isolatated persistence

  • No sneaking in to look at the other guys data.
  • Same team that writes the micorservice owns the data store OR - use a external data store
  • Still ISOLATED
  • Only external access to data store is via public interface

Event-triggered should be a first-class object#

  • A thing happend - Something I cared about
  • Asynchronous Operation - No care if someone is listening.
  • State changes or events must be used.
  • Why - Represents how the world works

Must be within an interface within microservices

  • interface is for any data into or out of a microservice
  • Stich Fix is still a mono data base with 175 tables
    • Single point of failure.

Solution is decoupling data how to do it #

  • Write a service interface to isolate data access for the table.
  • Rinse and repaet
  • Simple discussion but a lot of code changes and joins etc.

Bounded context mentioned many times.

Managing data within microservices.

  • shared data
  • Joins work well in mono-data

One service OWNS the data (Customer)

  • Every other service is a RO non-authoritative service

Approach one just look it up? - Every other service looks it up

If too expensive to do ...Eventing from Customer service

Shared metadata - Much of it does not change often.#

  • colors, US States, Shoe Sizes zip codes ....

Joins in microservices #

  • Approach one - Joins in application
  • make two calls to different services.
  • Works well for one-to-many just as a Web page does.

Approach two

  • Maintain a cache of a join
  • Item feedback example many-to-many
  • Listens to items service and feedback service to make join ....
  • Materializing a view in realtime
  • NoSql does this

Transactions #

Easy in relation db. ACID etc.
  • SAGA Pattern -
  • Commit - Workflow that updates different data stores.
  • Rollback - reverse workflow
  • Serverless - Functions as a Service is a great way to do this.

Big believer Lambda Google Cloud functions Azure Functions - triggered by events and produces an event

Stich Fix is hiring - 50% are remote

Events that do not arrive or like SLP exactly once deliver out of order more than once

Evens should use - At least once deliver

  • More than once Mutiple times - Item impotent
  • Out of order - Do you care - Consumer must maintain state
Create customer Delete Customer when out of order this is a probabilistic Cuncurrent replicated data types or Tomombstones

At most once - think of UDP

More Information#

There might be more information for this subject on one of the following: ...nobody

Add new attachment

Only authorized users are allowed to upload new attachments.
« This page (revision-2) was last changed on 04-Aug-2017 10:05 by jim