Ask HN: When do you choose RDBMS read replica over ElasticSearch and vice-versa?

One of my clients sync PostgreSQL DB (RDBMS) data with ElasticSearch. Because of DB overload, they have started migrating big SELECT queries (that have joins and aggregations in it) to APIs that read data from ElasticSearch.

Is this not an anti-pattern? Isn't it easier and nicer to read this data from PostgreSQL Read Replica instead of reading from ElasticSearch??