Re: tds_fdw : Severe performance degradation from postgresql 10.23 to 16.4

On 11/12/24 11:37, Achilleas Mantzios - cloud wrote:
>
> ...
> 
> We run perf on both systems for 90 seconds to make sure that it included
> at least two runs of the slow system, and we attach both outputs. We run
> perf as :
> 
> sudo perf record -g -p <PID> -- sleep 90
> 
> it strikes me that we dont find similarities between the two, but we are
> new to perf.
> 

Not sure, but it seems the slow profile has a lot of writes. Not sure
why. Do both instances have the same work_mem value / available memory,
storatge?

Maybe EXPLAIN VERBOSE would show if the remote query is the same ...

Maybe try setting log_temp_files=0 before running the query. Writing the
foreign scan result set into a temp file could be happening on one of
the machines only.


regards

-- 
Tomas Vondra