Andrew Dunstan <andrew@dunslane.net> writes:
> Tom Lane wrote:
>> ... which leads to the conclusion that those random little startup things
>> plperl does are just unbelievably expensive.
> I suspect much of this time is taken in loading perl modules, (strict,
> Safe, utf8, SPI among others), not in anything very special that we do.:
I did some more poking and determined that on my machine, it takes about
14 msec to do plperl_init_interp and about 20 msec to do
plperl_safe_init.
Most of the former time (over 9 msec) is going into the perl_parse call;
I'm not entirely sure why it's so expensive. Taking out the SPI module
initialization does *not* save anything worth noticing. I tried
trimming various bits of the PERLBOOT script and the only thing that
made a dent was removing "use vars qw(%_SHARED);", but that was still
not much (about 5 msec).
In plperl_safe_init, it's entirely clear that "require Safe" is the
expensive part; I can duplicate a comparable time expenditure when doing
that from the command line.
So about the only real answer is going to be preloading. It seems worth
considering that on machines where can_run_two is true, we should just
go ahead and initialize both interps at _PG_init time, so as to allow
the "require Safe" overhead to be bought back by preloading. I'm not
sure how to deal with the UTF8 hack though.
regards, tom lane