Decision to add relational database and choose PostgreSQL.

docs/architecture/decisions/0011-add-relational-database.md

+# 11. Add relational database
+
+Date: 2017-12-15
+
+## Status
+
+Accepted
+
+## Context
+
+Number of document types which does not use Elasticsearch's fulltext  
+capabilities is growing. Recently released Elasticsearch 6 is bringing one type  
+per index which means management of many indices.
+
+## Decision
+
+We will add relational database as primary database. Elasticsearch will be used  
+for denormalized reports and related data intended for fulltext search.
+
+## Consequences
+
+* Proper relations in data on database level.
+* Simplification in data management and schema migrations.
+* If ORM is used then GraphQL types may be derived without additional effort  
+and translations between document types and GraphQL types.
+* Less indices in Elasticsearch.

docs/architecture/decisions/0012-use-postgresql.md

+# 12. Use PostgreSQL
+
+Date: 2017-12-16
+
+## Status
+
+Accepted
+
+## Context
+
+We want to add relational database.
+
+## Decision
+
+We will use PostgreSQL. It's a mature database with handy features like JSON  
+and hstore data types. It's fully ACID compliant including schema changes. It  
+has very good support in Django's ORM.
+
+Another popular option is MySQL/MariaDB. But because it has a major bug `#28727`  
+(10 years since it has been reported and it's still not fixed) breaking ACID in  
+schema changes it can't be used for any serious project.
+
+## Consequences
+
+We will have mature relatonal database with handy JSON and hstore data types.