Also, treating json like a string is faster than treating it like an object. I'd be curious to see the Postgres numbers with the json in a varchar or similar.

yeah and in normal operation the OS/PG will have all your data in memory (or all your indexes if very, very much data or under spec'd db server).

Disk is for writes.

There is one disk operation that is guaranteed to happen when pushing data into a database like PostgreSQL. It will write data to an append-only binary log.

In Postgres this is called “write-ahead-log” or WAL. On MySQL/MariaDB it is called Binlog.

For a long time the only persistence redis even had was an append only binary log-style system.

Additionally you could tune redis to a maximum persistence threshold which emulated the absolute minimum persistence threshold of a system like Postgres or MySQL.

There is no comparison between these systems. One is designed to treat your data very safely by default and the other is designed for speed.

They both achieve those goals.

I don’t even want to go into the fact that vfs access isn’t still slower than direct memory access (for, I hope, many obvious reasons).

The topic of this thread is the equivalent of judging a fish by its ability to climb a tree.