The took time for creating 7000 records in an index with no duplicates is 450ms, when all are duplicates its 7000ms, a 16x performance decrease. Also I see large spikes in GC collection times. When I use index it's 450ms and when I use upsert it's 600ms.
This doesn't make sense, I'd think the op_type create would require the least processing time as it's not actually creating a record.
Using elasticsearch 7.3.1 with OpenJDK 11 on Ubuntu