Re: Incorrect accounting (n_tup_ins) of non-inserted rows

Ilya Matveychikov <matvejchikov(at)gmail(dot)com> writes:
> Seems that accounting of insertions with `n_tup_ins` not correct in
> case of insertion errors cause by constraints checking (duplicate key
> value violates unique constraint):

You already found one of the many code comments indicating that this is
the intended behavior:

> CODE REFERENCE (src/backend/postmaster/pgstat.c)
> /* count attempted actions regardless of commit/abort */

> So, is this behavior normal or probably needs to be fixed?

No, it's not a bug, and it's not going to be fixed. For many of the
intended applications of those counts (e.g, determining whether
autovacuum/autoanalyze is needed), this is the correct behavior
and ignoring actions of failed transactions would be incorrect.

The live/dead tuple counts do attempt to take transaction success
into account; perhaps looking at those would be more helpful for
your use-case?

regards, tom lane