i think i've encountered a bug in postgresql 8.1.
now i'm nearly positive it's a bug.
i created database in this way:
CREATE DATABASE leak;
\c leak
CREATE TABLE users (id serial PRIMARY KEY, username TEXT NOT NULL DEFAULT '', password TEXT);
then made a list of "usernames": from "aaaa" to "czzz" with probability 97% - 3% are missing.
and then i COPY'ied this list into users.
then:
CREATE UNIQUE INDEX xxx on users (username);
vacuum verbose analyze;
after all of this i run this script:
#!/usr/bin/perl
use strict;
use DBI;
my $dbh = DBI->connect("dbi:Pg:dbname=leak", "pgdba", "", {"AutoCommit" => 0,
"RaiseError" => 0, "PrintError" => 0}) or die "cannot connect\n";
for (1..1000) {
my $i = 0;
my $q = "zzz";
while (1) {
$dbh->rollback();
$dbh->do("INSERT INTO users (username) VALUES ('$q')");
$dbh->commit();
$q++;
last if $q eq 'fzzz';
$i++;
if (0 == $i % 1000) {
system("ps uxf");
}
}
print "one iteration done\n";
<>;
}
$dbh->disconnect();
technically - i think that after first "iteration" memory usage should not inreate. but it does.
it does in small amounts, but still does.
can anybody test/confirm the problem?
depesz
