2018-06-05 15:00 GMT+02:00 Andres Freund <and...@anarazel.de>: > Hi, > > On 2018-06-05 06:32:31 +0200, Pavel Stehule wrote: > > ./configure --with-libxml --enable-tap-tests --enable-debug --with-perl > > CFLAGS="-ggdb -Og -g3 -fno-omit-frame-pointer" > > > > [pavel@nemesis postgresql]$ gcc --version > > gcc (GCC) 8.1.1 20180502 (Red Hat 8.1.1-1) > > > > I executed simple script > > > > do $$ declare i bigint = 1; s bigint = 0; begin while i <= 100000000 > loop > > s := s + i; i := i + 1; end loop; raise notice '%', s; end $$; > > > > 7,68% postmaster postgres [.] > > GetSnapshotData ▒ > > 7,53% postmaster plpgsql.so [.] > > exec_eval_simple_expr ▒ > > 6,49% postmaster postgres [.] > > It seems to me the right fix here isn't a new class of functions, but > rather support for delaying the computation of the snapshot to the point > it's needed. That'll be far more generically applicable and doesn't > require user interaction. >
good idea. Can be quick fix. > > > > ExecInterpExpr ▒ > > 4,13% postmaster postgres [.] > > So we're going to need to optimize this further as well, I've a pending > patch for that, luckily ;) > nice :) > > LWLockRelease ▒ > > 4,12% postmaster postgres [.] > > That's also GetSnapshotData()... > there are about 10% locking, unlocking plan cache still. Regards Pavel > Greetings, > > Andres Freund >
