On SQL vs NoSQL
NoSQL is many products, not one architecture. Keep joins where the problem needs them; do not migrate wholesale on fashion.
“NoSQL” bundles dissimilar tools: document stores, wide-column / BigTable descendants, key-value caches, pure object stores. Cassandra is not MongoDB is not a column-oriented analytics engine.
Joins did not disappear
BigTable-style systems push joins into middle tiers or application code. Denormalising everything is a trade, not a free lunch — the same class of trade SQL shops already make with summary tables and caches.
Tool fit
SQL RDBMSs spent decades as the default general-purpose store. Other stores earned niches: write-heavy wide-column, flexible documents, graph-shaped walks (OQGRAPH). Moving an entire estate because a conference tagline changed is usually a fail. Pick the store for the access pattern; keep durable relational cores where constraints and ad-hoc joins still earn their keep.