A DBA (or at least I) would have an issue with referring to data combination by multiple PKs.

A proper solution could be to inject unique ID for each data row and operate on that. In cases of working on “database views” or other results of joins, no one can rely all the time on a combination of IDs to be unique because various join conditions can duplicate the records like there is not tomorrow and filters may obscure that.

I suggest that server (SQL server, or web app code) keeps the mapping of columns-data-combination-to-unique-id in ORM (object relational mapper) or on DB. jqGrid (in my tests) supports GUIDs (UUIDs) as row IDs quite fine. Every time I feed a “view” each row has a GUID assigned to it by the server. When “edit” “delete” comes back to it, it knows exactly what data combination (row in a view) this refers to.