If you need to retain the same order in your unique array as the original a different approach is required as hashes are inherently unordered. The grep uses the %seen hash to only pass through data elements that it hasn't already seen, thus removing duplicates from the stream. The first (lower) map takes each line read from the input file and removes any trailing white space, including line terminators (I do this instead of chomping in case there are differing numbers of trailing spaces in the data). The grep removes duplicates as already described and the second (upper) map creates an anonymous array of data split on commas.

$ perl -Mstrict -Mwarnings -MData::Dumper -E '
open my $inFH, q{<}, \ <<EOF or die $!;
server1,user1,%
server1,user1,db1
server1,user2,%
server1,user3,%
server1,user1,%
server1,user2,%
server1,user2,db2
server1,user3,db3
server1,user3,%
server2,user1,%
server2,user1,db1
server2,user2,%
server2,user3,%
server2,user1,%
server2,user2,%
server2,user2,db2
server2,user3,db3
server2,user3,%
server3,user1,%
server3,user1,db1
server3,user2,%
server3,user3,%
server3,user1,%
server3,user2,%
server3,user2,db2
server3,user3,db3
server3,user3,%
EOF

my @logins = do {
   my %seen;
   map  { [ split m{,} ] }
   grep { ! $seen{ $_ } ++ }
   map  { s{\s*$}{}; $_ }
   <$inFH>;
   };

print Data::Dumper->Dumpxs( [ \ @logins ], [ qw{ *logins } ] );'
@logins = (
            [
              'server1',
              'user1',
              '%'
            ],
            [
              'server1',
              'user1',
              'db1'
            ],
            [
              'server1',
              'user2',
              '%'
            ],
            [
              'server1',
              'user3',
              '%'
            ],
            [
              'server1',
              'user2',
              'db2'
            ],
            [
              'server1',
              'user3',
              'db3'
            ],
            [
              'server2',
              'user1',
              '%'
            ],
            [
              'server2',
              'user1',
              'db1'
            ],
            [
              'server2',
              'user2',
              '%'
            ],
            [
              'server2',
              'user3',
              '%'
            ],
            [
              'server2',
              'user2',
              'db2'
            ],
            [
              'server2',
              'user3',
              'db3'
            ],
            [
              'server3',
              'user1',
              '%'
            ],
            [
              'server3',
              'user1',
              'db1'
            ],
            [
              'server3',
              'user2',
              '%'
            ],
            [
              'server3',
              'user3',
              '%'
            ],
            [
              'server3',
              'user2',
              'db2'
            ],
            [
              'server3',
              'user3',
              'db3'
            ]
          );
$
[download]

I hope this is of interest.