SQL is set-oriented. How would that work on a potentially indefinite stream, other than as a simple filter which you could just do with a tool such as awk.
Many relational operations don't require a whole stream to compute, and many of those that do don't need it all at once.
Projection (mapping), a join against a fully loaded other side as well as filtering work.
Aggregation can consume an indefinite stream with limited working set if the cardinality of the grouping key isn't large.
And of course you can combine these in nested and unioned operations, computing across multiple indefinite streams concurrently and with limited working set.
It would be tricky to make work effecively without hinting for things like joins, for sure; join order is one of the hardest bits a query engine optimizes.
Projection (mapping), a join against a fully loaded other side as well as filtering work.
Aggregation can consume an indefinite stream with limited working set if the cardinality of the grouping key isn't large.
And of course you can combine these in nested and unioned operations, computing across multiple indefinite streams concurrently and with limited working set.
It would be tricky to make work effecively without hinting for things like joins, for sure; join order is one of the hardest bits a query engine optimizes.