I've noticed that commercial databases (e.g. SQLServer) and some open source databases (e.g. Cubrid) support so called "bookmarks". As far as I understood, a bookmark allows quickly jump to a row for retrieval or modification.
Here is scenario that I'm trying to deal with: A BI/ETL application is querying the Postgres database. The queries return lots of rows (36 mil), each is about 1KB or larger. So I'm using DECLARE CURSOR/FETCH to read the rows into a buffer (size is 10000 rows, and I'm freeing memory for oldest rows). The application may alter or re-read some previously read rows by the row index. Problem is: if a row is not in the buffer (freed) the application cannot resolve row index into row itself. I considered using a unique key to located the row, but unfortunately some queries do no allows determining the most unique key.
I'm thinking, is it possible to retrieve/alter row by its index after a Postgres Cursor have read the row?
The application allows a customer to define DB Schema as well as the queries, so my code does not have a prior knowledge about DB and queries. It is supposed to provide a certain API with functions based on row indexes. The API was initially designed for SQLServer, so the goal is to migrate the application from SQLServer to Postgres.
