No, If possible, match it in Oracle, your (left/right) joins with those big table(s). Not in Talend tMap.
Your DB engine is designed to perform these kind of tasks...
Talend/java job will eventually crash if there's no memory left... inevitable heap space shizzle
So your matching strategy should be.... (if you insist on Talend doing the matching)
0 - only get data which you need to match. Reduce data footprint and calculate max needed heap space.
1 - Make sure Cartesian products don't occur by mistake, if it should be 1:n and it turns out to be n:n... check for duplicates on the unique side
2 - Match on relevant columns ... store match.... and match it back and add those needed additional columns. Reduction of bytes!
Never used it but I will in the future, mem analyzer... http://www.eclipse.org/mat/
However if a virus program runs in the background it could prevent allocating memory above 512MB... switch it off. And finally dont get into garbage collection stuff, your jvm does it for you,
Min heap space is to increase startup time... but Max heap space is what is max allowed by jvm, so focus on your max value calculation. Some processes need so much memory you can only run it separately ... probably this 4mln matching records, its the elephant in the room.
