Compare

Side-by-side on the decisions that recur: index vs scan, normalize vs denormalize, optimistic vs pessimistic, partition vs shard, and more — with when to choose each.

Index Scan vs Sequential Scan Normalize vs Denormalize Optimistic locking vs Pessimistic locking Read Committed vs Serializable Partitioning vs Sharding Read replica vs Shard Relational vs Document Postgres + pgvector vs Dedicated vector DB Cache-aside vs Write-through B+ tree storage engine vs LSM tree storage engine Heap table + secondary indexes (PostgreSQL-style) vs Clustered primary index (InnoDB-style)LRU vs Clock (second chance)Synchronous replication vs Asynchronous replication

	Postgres + pgvector	Dedicated vector DB
Systems to run	One	Two (plus a sync boundary)
Filters	SQL, transactional with the data	Engine-specific; joins done in your code
Scale ceiling	Tens of millions of vectors	Hundreds of millions to billions
Consistency	Same transaction as the rows	Eventual with the source of truth
When	Vectors alongside existing relational data	Vector search is the core workload at proven scale
Choose this when	The default for RAG and recommendations — start here; the vectors live next to the rows they describe.	When a measured number shows pgvector cannot keep up and vector search is the product.