This document outlines best practices for building enterprise Shiny applications. It recommends starting with a prototype to get customer buy-in before productionizing. For maintainability, developers should use modules, extract logic, and load data from databases. The document also stresses the importance of testing with a pyramid approach, avoiding errors in production through data validation, and scaling applications through load balancing and performance profiling. Fully automating deployment, tests, and code quality checks is also advised.

1. Best practices for building Shiny enterprise applications / useR! 2019
Toulouse
Filip Stachura @filipstachura

2. OUR SERVICES
50
Commercial Shiny
dashboards
delivered
Projects delivered
700
Max no. of concurrent
users using our Shiny
dashboard
40
● Data Science consulting
● Business applications of
Machine Learning
● Satellite data analysis
● R & Python programming
● World class R-Shiny
production solutions
@filipstachura

3. OUR WORK
@filipstachura@filipstachura

4. SELECTED CLIENTS
@filipstachura

5. Why ?
@filipstachura

6. WHY SHINY
Django, React,
Angular, Go, Scala...
Fast prototyping.
One language.
Excel. Clean structure.
Scalability. Correctness.
Business Intelligence.
Tableau, PowerBi, Qlik
No limitations.
Full power.
@filipstachura

7. 1
Start with prototype.
@filipstachura
6 LESSONS FOR ENTERPRISE SHINY
2
Build for maintainability
and dev velocity.
4
Avoid errors, learn
from mistakes.
3
Create a test
pyramid.
5
Scale and perform.
6
Automate.
Automate fully.

8. START WITH PROTOTYPE
Customer
Development
Buy-in
from users
Productionization
@filipstachura

9. Traits of a good architecture:
1. Use of modules.
2. Logic extracted and tested.
3. No heavy computations*.
4. Data loaded from database.
MAINTAINABILITY AND VELOCITY
* but avoid premature optimization!
@filipstachura

10. Source: https://watirmelon.blog/testing-pyramids/
SHINY TEST PYRAMID
@filipstachura
Ideal Software Testing Pyramid

11. Source: https://watirmelon.blog/testing-pyramids/
SHINY TEST CONE
@filipstachura
Software Testing Ice-cream Cone Anti-Pattern

12. 1. Always validate data.
2. Dataset owner alerts.
3. Logging infrastructure.
4. Fault tolerance.
@filipstachura
AVOID ERRORS IN PRODUCTION

13. AVOID ERRORS IN PRODUCTION
@filipstachura
Daily data validation email
Style improvement
Continuous integration
Show data status in app

14. Read more: appsilon.com/data-quality/
DATA VALIDATION LEVELS Data quality is more than structure and format
@filipstachura

15. 1. Load balance:
a. RStudio Connect
b. Amazon ELB
2. Profile.
3. Run performance tests.
c. headless
d. non-headless
@filipstachura
SCALING AND PERFORMANCE

16. @filipstachura
EXAMPLE ARCHITECTURE
MASTER
NODE
ANSIBLE R
(RESULTS PROCESSING+VIS)
QUERY
SERVER1
QS2
QS3
QSn
GOOGLE
PUPPETEER
LOAD BALANCER
AWS APPLICATION
LOAD BALANCER
APP1
APP2
APPn
DB

17. DEPLOYMENT STRATEGIES
1 deployment
environment.
Manual.
Not optimized.
2-3 environments: staging,
test, production.
Automated (e.g. ansible, docker)
saves time, easy roll back, one click to
drop or create server
Avoid static files. Bundle
what you can.
@filipstachura

18. UI & FEATURES
AUTOMATED TESTS
DATA VALIDATION IN ETL
LOAD & PERFORMANCE TESTS
SSO INTEGRATION
AUTOMATED DEPLOYMENT
END-TO-END TESTS
AUTOMATED CODE QUALITY CHECKS
DECISION SUPPORT SYSTEM: WHAT’S BELOW THE SURFACE?
@filipstachura

19. Shiny is ready for production!
appsilon.com
filip@appsilon.com
Let’s grab a coffee! ☕