PostgreSQL metrics disparity between query time and block read/write time

As we discovered in  PMM-5698  + puneet.kala investigation with  PG team : 

We have metrics in PG and they are cumulative but by the data, we have in QAN - it looks like we treated them as data for current query or something like this  

• 'blk_read_time',
• 'blk_write_time',

 

Because we had 'blk_*_time' metrics bigger than 'query time' 

 

STR:
1) Setup PostgreSQL 10.4 version and "create extension pg_stat_statements"
2) Create database sbtest_pg
3) run alter database sbtest_pg set track_io_timings=1;
4) create user sbtest
5) grant the user all privileges to sbtest_pg
6) sysbench --db-driver=pgsql --oltp-table-size=100000 --oltp-tables-count=24 --threads=1 --pgsql-db=sbtest_pg --pgsql-user=sbtest --pgsql-host=localhost /usr/share/sysbench/tests/include/oltp_legacy/parallel_prepare.lua run
7) sysbench --db-driver=pgsql --report-interval=2 --oltp-table-size=100000 --oltp-tables-count=24 --threads=32 --time=60 --pgsql-host=localhost --pgsql-port=5432 --pgsql-user=sbtest --pgsql-db=sbtest_pg /usr/share/sysbench/tests/include/oltp_legacy/oltp.lua run