I've been periodically collecting the stats stored in
pg_statio_all_tables and pg_stat_database for ~30 different
clusters, and have noticed a curiosity.
I would have thought that for a given period, the change in
pg_stat_database.blks_read would be <= the sum of the changes in
pg_statio_user_tables.heap_blks_read +
pg_statio_user_tables.idx_blks_read +
pg_statio_user_tables.toast_blks_read +
pg_statio_user_tables.tidx_blks_read.
In short, the total would be <= heap + idx + toast + idx for user
tables.
It does not appear that way. The table-level IO stats appear to
be typically 1-2 orders of magnitude larger than the db-level
stats. Can anyone explain that?
TIA.
Ed