A presentation on clearly defining a microservice architecture, culture, and discovering how to determine whether it is a step in the right direction for your system. I discuss about the decisions that lead us to take on a microservice architecture approach at Sprout, and the challenges we are facing as a startup learning a new method for making apps.

1. Refactoring to Microservices
Moving away from old school servers
and into the cloud

2. ASK AWAY!
Step 1. Go to slido.com
Step 2. Key in #SproutAcademy in the event code.
Step 3. Type your questions.
Wifi name: KMC
Wifi password: S0lstic3!

3. About Me
Senior Architect at Sprout
Solutions
Microsoft MVP for Visual Studio
and Development Technologies
User Group Lead of the
Philippine .NET Users Group
(PHINUG)

4. Not so long ago in a country far far away...

5. In January 2009 a startup purchased 3 servers...
Web Database
2 servers 1 server
4 cores, 2.83 GHz, 12 MB L2
cache
8 cores, 2.5 GHz, 24 MB L2
cache
8 GB RAM 24 GB RAM
500 GB RAID 1 mirror array
(hot-fail, cold-swap)
500 GB RAID 10 array
(hot-swap, up to 2 drives at once)
dual redundant power supplies
Total Cost: ~$5500 (one time) + colocation fee

6. Had to expand by December...
January 2009
December 2009

7. Seven years later...
Web Main Database Everything else
DB
And a bunch of
other servers...
11 servers 2 servers 2 servers ● Service Servers
● Redis Servers
● Elasticsearch
Servers
● HAProxy Load
Balancer Servers
● Machine Learning
Servers
● Logstash Servers
● HTTP Logging
SQL Server
12 cores @2.6-3.5 GHz 12 cores @2.7-3.5 GHz 8 cores @3.2-3.6GHz
64 GB RAM 384 GB RAM 768 GB RAM
2x 300GB SATA SSDs
RAID 1
4 TB PCIe SSD (RAID 0)
+ 24 x 200 GB SATA
SSDs (RAID 10)
3x 2 TB PCIe SSD (Raid
0) + 24 x 1.2 TB SATA
HDDs (RAID 10)
Dual 10Gbps network Dual 10Gbps network Dual 10 Gbps network
Total Cost: $$$$$$$$$$$$

8. And their data center looks like...

9. Statistics for 2016
209,420,973 HTTP requests to our load balancer
66,294,789 page loads
504,816,843 SQL Queries (from HTTP requests alone)
5,831,683,114 Redis hits
17,158,874 Elastic searches
607,073,066 ms (168 hours) spent running SQL queries
10,396,073 ms (2.8 hours) spent on Redis hits
1,609,944,301 ms (447 hours) spent processing in ASP.Net

10. That’s the story of...

12. Highly performant apps can be
achieved with well designed
monoliths

13. Monoliths are not necessarily bad
Simplicity
Consistency
Inter-module refactoring

14. So why would you use microservices?

15. Let’s talk about

16. Sprout also begins its story in 2009 with...

17. Typical monolithic WISA application except...
ASP.NET Web Application
Written in C#
Runs on IIS
SQL Server Database
Inherently multi-tenant

18. Then came along...

19. Integration became a business need...
Biometric data upload
Data integration between apps

20. More clients & users meant
more business needs came along...
Biometric data upload
Data integration between apps
Clients need APIs
High Availability
Performance
New products
Tangential features
Better UX

21. Microservices should answer concrete
business needs

22. Monolithic app vs. Microservice apps

23. You have a true microservice when you achieve...
Componentization via services
Organized around business
capabilities
Products not Projects
Smart endpoints vs. dumb pipes
Decentralized governance
Decentralized data management
Infrastructure automation
Design for failure
Evolutionary Design

24. Componentization via services
Payrol
l
HR
● Authentication
● Employee 201
● Attendance
● Performance
● Movement
● Compliance
● Loans
● Authentication
● Attendance
● Taxation
● Loans
● Accounting
● Compliance

25. Componentization via services
Payrol
l
HR
Employee 201 Attendance
Performance
Movement Compliance
Loans
Accounting
Taxation
Authentication

26. Organized around business capabilities
Each team must be full stack
Developer
QA
UI
DBA
System Admins/Devs
} DevOps culture

27. Products not projects
Each team must have a dedicated product owner
Preferably a domain expert
Otherwise needs to have immersed with end users
Team should be concerned over the product lifetime

28. Refactoring to microservices means
reorganizing your team

29. Smart endpoints vs. dumb pipes
Endpoints should have full business logic
Security
Data validation
Error handling
Logging
Do not make CRUD endpoints

30. Decentralized governance & data management
Each team must have a say on their development stack
UI/UX
Backend
QA
Each team must have a say on their data store
SQL vs. No SQL?
Must otherwise be compliant with business requirements

31. Infrastructure automation
Capability to quickly provision servers
Quickly scale based on need
Ability to deploy applications rapidly

32. Design for failure
Gracefully handle errors
Rapidly provide feedback on errors
Logging and monitoring

33. Evolutionary Design
Services should be able to change without breaking
Handle both old and new endpoints
Deprecate over time

34. Microservices have their own benefits
Partial Deployment
High Availability
Preserve Modularity
Multiple Platforms

35. Our microservices journey so far

36. Microservices across products
PayrollHR Project
X
● Authentication
● Employee 201
● Attendance
● Performance
● Movement
● Compliance
● Loans
● Authentication
● Attendance
● Taxation
● Loans
● Accounting
● Compliance
Employee 201

37. Challenges: Infrastructure
Web
Live
Data
Live
Web
Test
Data
Test
vs

38. Option 1: On Premise
Web
Live
Data
Live
Web
Test
Data
Test
● Pros: Greater degree of
control, more cost
effective (up to a certain
point)
● Cons: More difficult to
expand and provision,
less resilient to downtime,
need relevant skillset

39. Option 2: Public Cloud (AWS / Azure / Google)
● Pros: Faster provisioning,
more resilient, less
management overhead
● Cons: Beefed up specs
and bandwidth are
costlier, hidden costs,
data residency
complications

40. Challenges: Data architecture
Employee 201Sprout HR Sprout Payroll
Sharded by tenant Sharded by tenant
New!

41. Option 1: Unsharded data
Sprout HR Sprout Payroll
Sharded by tenant Sharded by tenant
● Pros: Simpler data access,
easier maintenance for
backups & business
continuity
● Cons: Lost data isolation,
potential bottlenecks at
data layer
Employee 201

42. Option 2: Aggressive sharding
Sprout HR Sprout Payroll
Sharded by tenant and by
employee table
Sharded by tenant
● Pros: Greater data
isolation and modularity;
Shards could be deployed
to separate servers for
better throughput
● Cons: More difficult
maintenance for backups
& business continuity
Employee 201

43. Our journey towards...
Componentization via services
Organized around business
capabilities
Products not Projects
Smart endpoints vs. dumb pipes
Decentralized governance
Decentralized data management
Infrastructure automation
Design for failure
Evolutionary Design

44. Will be continued...

45. Panel / Q&A
Thank you very much!

46. References
https://stackoverflow.blog/2009/01/12/new-stack-overflow-server-glamour-shots/
https://nickcraver.com/blog/2016/02/17/stack-overflow-the-architecture-2016-
edition/
http://blog.serverfault.com/2015/03/05/how-we-upgrade-a-live-data-center/
https://martinfowler.com/articles/microservices.html
GOTO 2014 - Microservices - Martin Fowler https://youtu.be/wgdBVIX9ifA