John Coers <coers@intrinsity.com> writes:
> I've attached a little facsinating truss output.
You probably shouldn't have sent 250k of trace to the whole mailing
list. But it *is* fascinating.
I tried to duplicate the result locally with no success. For a
10000-row COPY FROM STDIN (using the tenk1 table from the regression
database, which has several indexes), my trace contains:
lseek: 14060
read: 4261
write: 4967
recv: 43
other: 170
with a large part of the "other" being process startup/shutdown
overhead. There are no semops or setitimers at all in my trace.
There do seem to be some excess lseeks, but I don't see lots and
lots of seeks with no read or write as you show.
Could we see the schema for the table you are using?
("pg_dump -s -t tablename dbname" is the best way to show it.)
regards, tom lane