How to catch a request floating in time?

Question

PostgreSQL

There is a request floating on time.
Most of the time it runs in milliseconds, but occasionally there are dips of up to 6 seconds.
After, with the parameters that worked out 3-6 seconds, the request is executed at a normal speed in ms

. How can you understand what is happening?

There are no changes to the processor at this time.
On the disks, there is a very small drawdown in time.
There are no peaks in memory either.

Auto_explain is enabled, but we can't enable it on the analyze level because it will slow down the database a lot.
I still don't quite understand what to do

Answer 1

After, with the parameters that worked out 3-6 seconds, the request is executed at a normal speed in ms

Symptoms of slow reading from disks.
Set pg_stat_statements, enable track_io_timing and look at the blk_read_time, blk_write_time of this request for the number of calls and the total time they were executed.
PS: Please note that the latency of the disk system is critical for the DBMS. Are there io latency graphs for reading and writing?