I don't think it has to do with that. It's fine to pull chunks of data (by a small number of rows, say, 4000), so long as we expose them as an iterator and discard the processed ones from memory. The problem is that they stay in memory.

Are you closing the cursors?

Oh yes.

I can't use with-open, though. The cursors are closed inside .onComplete of the Scala future returned when the flow graph begins to run.

Otherwise, they'd be closed before the flow is done.

https://github.com/funcool/clojure.jdbc/issues/44

Huh, interesting. Thanks!

Hmm, but opts in my code above is actually {:fetch-size 4000}, anyway.

Just need auto-commit.

basic question — in next-jdbc can i express somehow that i want to get back long instead of BigDecimal? eg.

(jdbc/execute! (jdbc/get-connection {:dbtype "oracle" ,,, }) ["select 1 from dual"])
=> [{:1 1M}]

Perhaps working with might shed some light?

I don't work with any oracle db's but I suppose controlling how next jdbc converts each column into the appropriate type could be handled here.