On Mon Mar 23, 2026 at 9:50 AM CET, Jelte Fennema-Nio wrote:
I think Andres' reply also applied to that suggestion. Partitioning really messes up jit for expressions too. I was able to pretty easily create some bad cases even with jit_tuple_deforming set to off. A query involving many partitions, but with relatively few rows per partition, jit became ~54x slower when using -O2 (743ms vs 40318ms). And even with -O0, jit is still ~3x slower (221ms vs 711ms). Full script and results attached.
Resending the above with a compressed attachment, because it was too large for the mailing list.
jit_benchmark.sql
Description: application/sql
results.out.gz
Description: application/gzip
