Hey Jeff,
First of all thank you again for the quick response. I really appreciate your comments.
Unfortunately I installed pg from rpms so I cant compile my current env but I can install it from source code and migrate the data via pg_dump. Can you explain how can I compile the sources without this commit ?
I understand that my test such a good test but I thought that using something simple can show my issue. I was sure that I added the results of the same tests after disabling the param but it seems that I didnt do it. I'll send it tomorrow because right now im not inforont of my pc. The bottom line, after disabling this param the same query took less time in pg12(significly).
From this email chain and other peoples comments it seems that I'm not the only one who was facing this issue. One thing that might worth mentitoning, in all my tests I didnt use a big env, my db was at max 30GB and the max duration of a query was about 10s. It seems that this happens when we use queries with short duration (in your test you mentioned u used the same test but with 100 times more rows, maybe that is the reason it took more time ?) - maybe on "short" queries it creates degredation ? It is just an assumption...
I'll try to do most tests tomorrow and update with results.