This document discusses polyglot persistence, which is using multiple specialized databases rather than a single general-purpose database. It provides examples of VidaXL's use of polyglot persistence, including MySQL, MariaDB, PostgreSQL, SOLR, Elasticsearch, MongoDB, Couchbase, and Prometheus. The benefits discussed are using the right database for each job and gaining flexibility as the company transitioned to microservices. Challenges included increased complexity, and solutions involved automation, tooling, and hiring database experts.
Polyglot Persistence at VidaXL: Using Multiple Databases
1. Art van Scheppingen - Senior (No)SQL DBA @ VidaXL
Bart Oleś - Senior Support Engineer @ Severalnines
Polyglot persistence: utilizing
open source databases as a
Swiss pocket knife
3. VidaXL Company Stats
• Online retailer in (mostly) slow moving goods
• Founded 2008
• 350M turnover, 40% growth yearly
• 1500 employees (US, CN, AU, IN, RO, UA)
• HQ in the Netherlands
• 4 warehouses worldwide (NL, US and AU)
4. How does VidaXL sell its goods?
• Own webshop platform in EU, US and AU
• Warehouses in NL, US and AU
• Selling on other platforms, e.g. Amazon, eBay
• Allow selling on our own platform using Mirakl
• B2B drop-shipments
5. VidaXL Technical Foundations
• SAP as ERP system
• Genesys as CS system
• Webshop
• Open source web-based development strategy
• PHP / NodeJS
• Docker
• Cloudflare workers
7. What is Polyglot Persistence?
Using multiple specialized persistent stores rather than one single general-purpose database
8. Where does the term come from?
• The way we work is changing
• Enterprise applications are becoming more complex
• Separate (devops/agile) teams
• Ownership of applications
• (Micro)services
• Everyone has their preference
• Various programming languages
• Various storage systems
9. Where does the term come from?
• Monoglot Programming
• Only one programming language allowed
• Readability
• All code is in the same language
• Support
• One platform to support
• Knowledge
• Everybody is an expert
• Is there a jack-of-all-trades language?
10. Where does the term come from?
• Monoglot Programming
• Only one programming language allowed
• Readability
• All code is in the same language
• Support
• One platform to support
• Knowledge
• Everybody is an expert
• Is there a jack-of-all-trades language?
13. Polyglot Programming
• Polyglot Programming
• Use programming languages for what they are good at
• Flexibility
• Use Java for a secure API
• Use Scala for real time stream processing
• Use Python for text analysis
• Tie everything together using AngularJS
• Knowledge
• Everybody is expert at one or more languages
16. Data storage landscape changes
• Relational data stores (RDBMS)
• Key-Value data stores (“NoSQL”)
• Columnar data stores (OLAP)
• Document data stores (NoSQL)
• Graph data stores (GDB)
• Big Data
17. Data storage landscape changes
Software
RDBMS Oracle, MySQL,
PostgreSQL
Key-Value Redis, Riak
Columnar InfiniDB, Clickhouse
Document MongoDB,
Couchbase
Graph Neo4J, Janusgraph
Big Data Hadoop
20. Polyglot Persistence
• Complex problems require different storage systems
• Use the right tool for the job, for example
• Use PostgreSQL for financial data
• Use MySQL for website contents
• Use MongoDB for user profiles
• Use Cassandra for real time streams
• Use Neo4J for recommendation analysis
21. Use the right tool for the right job
Document storage: MongoDB
22. Use the right tool for the right job
Columnar storage: Cassandra
23. Use the right tool for the right job
Graph storage: Neo4J
26. Quick recap on our data stores
• MySQL
• MariaDB (Galera) clusters
• MySQL replication
• ProxySQL
• PostgreSQL
• SOLR
• Elasticsearch
• ELK
• MongoDB
• Couchbase
• (RabbitMQ)
• Prometheus
27. How did this happen?
• Continuous growth
• Hardly any time to overhaul existing systems
• Transition from monolith to microservice architecture
• For each microservice the most optimal solution has been chosen
• Early adopters of new technology
• Gaining advantage over competition
34. What were the challenges?
• Automation
• Increased complexity
• Systems monitoring
• Multiple integrations
• Maintenance becomes more difficult
• Backups
• Scaling
• Software updates
• DevOps are not a DBA
35. What were the solutions?
• Invest in automation
• Never perform any (large) task thrice
• Increase tooling
• Build it ourselves costs time
• Buying/licensing tools costs money
• Keeping the headcount low saves money
• Focus on systems that matter most
• Get (exteneral) help
• Hire DBAs! ;)